Inventory monitoring system for precious metal gold material

Abstract

The invention discloses an inventory monitoring system for precious metal gold materials, and relates to the technical field of inventory monitoring, and the system comprises the following components: a sensor node unit, a group intelligent collaborative network module, a central processing unit, a database unit and a human-computer interaction interface. According to the method, multiple Internet of Things sensor nodes are deployed, the dynamic threshold value adaptive algorithm is adopted to perform preliminary analysis and judgment on the collected data, and the threshold value can be dynamically adjusted according to the environmental change, so that the accuracy of abnormality judgment is improved, misjudgment or missed judgment caused by environmental gradual change is avoided, and the accuracy of abnormality judgment is improved. The group intelligent collaborative network module sends signals and data to peripheral nodes for collaborative verification when the nodes detect abnormal data, the credibility of each node is comprehensively considered by adopting a credibility weighted voting algorithm, the reliability of collaborative verification is improved, and when the sensor nodes send abnormal signals, the reliability of collaborative verification is improved. And a transmission path is selected by adopting a dynamic routing optimization algorithm, so that the abnormal signal can be quickly and stably transmitted.

Description

Technical Field

[0001] This invention relates to the field of inventory monitoring technology, specifically to an inventory monitoring system for precious metals. Background Technology

[0002] With the continuous development of the precious metals market and the widespread application of precious metal materials in industry, investment and other fields, higher requirements have been placed on the accuracy and security of their inventory management. Precious metal materials, such as gold and silver, have become indispensable materials in many industries due to their high value and easy circulation. However, the inventory management of precious metal materials faces many challenges, including the impact of environmental factors on the quality of gold materials, the risk of theft, and inaccurate inventory data.

[0003] Traditional methods for monitoring precious metal gold inventory mainly rely on manual inspections and simple sensor devices, which have many limitations. First, manual inspections are not only inefficient but also difficult to conduct 24-hour uninterrupted monitoring, easily leading to safety hazards. Second, although simple sensor devices can achieve basic data collection, they often lack intelligent analysis functions and cannot dynamically adjust monitoring thresholds according to environmental changes, easily resulting in false alarms or missed alarms. In addition, traditional systems also have shortcomings in data transmission and processing. The transmission of abnormal signals may be delayed due to improper path selection, affecting the timeliness of anomaly detection. At the same time, the lack of a collaborative verification mechanism means that the failure of a single node may lead to false alarms or missed alarms for the entire system, reducing the reliability of monitoring.

[0004] Given the shortcomings of traditional precious metal inventory monitoring technologies, it is of particular importance to develop an inventory monitoring system for precious metals. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an inventory monitoring system for precious metal gold. It can achieve real-time monitoring and accurate judgment of the gold storage environment by deploying multiple Internet of Things sensor nodes and adopting a dynamic threshold adaptive algorithm. At the same time, the introduction of a swarm intelligence collaborative network module improves the reliability of collaborative verification of abnormal data, and the use of a dynamic routing optimization algorithm ensures the fast and stable transmission of abnormal signals.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an inventory monitoring system for precious metal gold materials, the system comprising the following components: a sensor node unit, a swarm intelligence collaborative network module, a central processing unit and a database unit, and a human-computer interaction interface;

[0007] The sensor node unit: Multiple Internet of Things sensor nodes are deployed in the precious metal storage area. Each node has a built-in microprocessor, communication module and storage module, which are used to collect data on the gold storage environment and perform preliminary analysis and judgment.

[0008] The swarm intelligence collaborative network module connects each sensor node through wireless communication technology. When a node detects abnormal data, it sends signals and data to surrounding nodes, and the surrounding nodes perform collaborative verification.

[0009] The central processing unit communicates with the sensor node unit and the swarm intelligence collaborative network module, receives, stores and analyzes data, processes abnormal data after collaborative verification and generates early warning information, and manages and configures the network.

[0010] The database unit stores historical data, threshold ranges, abnormal event records, and node configuration information collected by the sensor nodes.

[0011] The human-computer interaction interface allows managers to view monitoring data, early warning information, and historical records, and to set system parameters and configure functions.

[0012] Furthermore, in the sensor node unit, each sensor node employs a dynamic threshold adaptive algorithm for preliminary data analysis and judgment. The algorithm formula is as follows:

[0013]

[0014] Where T new For the updated threshold, T old The original threshold is α, which is an adaptive adjustment coefficient ranging from 0.1 to 0.5. This coefficient is determined by the system through machine learning algorithms based on historical data fluctuations. When data fluctuations are large, α takes a larger value, and vice versa. i For the data collected in the i-th time, This is the average value of the most recent n collected data, where n is the data statistical period, set according to the sensor type. For example, n=10 for temperature and humidity sensors and n=5 for weight sensors. Through this algorithm, sensor nodes can dynamically adjust the threshold according to environmental changes, improving the accuracy of anomaly detection and avoiding misjudgments or missed judgments caused by gradual environmental changes.

[0015] Furthermore, when analyzing images, the image sensor in the sensor node unit employs an anomaly detection method based on feature pyramids and an attention mechanism. This method first extracts multi-scale features from the image through a feature pyramid network to obtain feature maps at different levels. Then, an attention mechanism is introduced to assign corresponding weights to each feature map to highlight key information. The weighted feature maps are then fused to form the final feature representation. Finally, a classifier and a regressor are used to analyze the final features to determine whether there are any anomalies in the image and to identify the location and category of the targets. The parameters of the attention mechanism are obtained through training on a large amount of labeled gold storage scene image data. The classifier and regressor are trained using a combination of cross-entropy loss function and mean squared error loss function, thereby effectively improving the detection accuracy and speed of the image sensor for anomalies in the gold storage environment and avoiding security risks caused by missing small anomalies.

[0016] Furthermore, in the aforementioned swarm intelligence collaborative network module, the collaborative verification between nodes employs a credibility-weighted voting algorithm. When node A detects an anomaly, it notifies its neighboring nodes B1, B2, ..., B... m After sending the verification request, each node makes a judgment based on its own detection data. The final anomaly judgment result R is determined by the following formula:

[0017]

[0018] Where w j For node B j The credibility weight, ranging from 0 to 1, is determined by a comprehensive evaluation based on factors such as the accuracy of historical data and device health. For example, a node with a historical data accuracy rate of over 95% has a higher credibility weight. j For nodes with values ​​between 0.8 and 1, and an accuracy rate of 80%-95%, w j The value of r is between 0.5 and 0.8. j For node B j The judgment result is given by m, which is the number of surrounding nodes participating in collaborative verification, and β, which is the voting threshold, ranging from 0.5 to 0.8. The threshold is determined by the system through simulation experiments based on actual application scenarios and historical collaborative verification results. This algorithm can comprehensively consider the credibility of each node, improve the reliability of collaborative verification, and prevent incorrect judgments caused by individual faulty nodes.

[0019] Furthermore, in the aforementioned swarm intelligence collaborative network module, when a sensor node sends an abnormal signal, a dynamic routing optimization algorithm is used to select the transmission path. This algorithm is an improvement based on the ant colony algorithm, and the formula is as follows:

[0020] τ ij (t+1)=(1-ρ)τ ij (t)+Δτ ij (t)

[0021] Where τ ij (t+1) represents the pheromone concentration on the path from node i to node j at time t, ρ is the pheromone evaporation coefficient, ranging from 0.1 to 0.3, determined based on network environment stability; it takes a smaller value when the environment is stable and a larger value when the environment changes frequently, Δτ ij (t) represents the increment of pheromone along the path from node i to node j at time t, Δτ ij (t) represents the pheromone increment on the path from node i to node j at time t. The pheromone increment is calculated using the following formula:

[0022]

[0023] Where Q represents the total amount of pheromones released, determined according to the urgency of the anomaly; the higher the urgency, the larger the Q value. k Let be the length of the path traversed by the k-th ant. This algorithm can dynamically select the optimal transmission path based on network node status, signal transmission quality, etc., to ensure the rapid and stable transmission of abnormal signals and improve the timeliness of collaborative verification.

[0024] Furthermore, when the central processing unit further analyzes the abnormal data, it employs an anomaly severity assessment algorithm, the formula of which is:

[0025]

[0026] Where S represents the severity score of the abnormality, and v k Let c be the weight of the k-th abnormal factor, determined based on the degree of influence of the abnormality type on the precious metal gold. For example, the weight of gold theft is 0.8, and the weight of slight temperature and humidity exceedances is 0.2. k S is the quantitative value of the kth abnormal factor, and s is the number of abnormal factors. Based on the calculated S value, the central processing unit generates different levels of early warning information according to different score ranges, such as S≥80 for severe abnormality, 60-79 for moderately severe abnormality, 40-59 for general abnormality, and S<40 for minor abnormality. It also adopts different notification methods and processing procedures to achieve precise hierarchical management of abnormalities.

[0027] Furthermore, when optimizing the detection rules of the sensor nodes, the central processing unit employs a reinforcement learning-genetic algorithm fusion optimization algorithm. First, the reinforcement learning algorithm is used to allow the sensor nodes to perform data detection and decision-making in different simulated environments, receiving rewards or penalties based on the decision results, and continuously adjusting the detection rules to maximize long-term rewards. Then, a genetic algorithm is introduced to evolve and optimize the rules obtained from reinforcement learning through selection, crossover, and mutation operations, using the following formula:

[0028] R new =GA(RL(R)old ))

[0029] Where R new For the optimized detection rules, R old The initial detection rules are defined by RL (Reinforcement Learning) and GA (Genetic Algorithm). In the algorithm, the reward function of reinforcement learning is set according to the anomaly detection accuracy and false alarm rate indicators, and the crossover probability and mutation probability of the genetic algorithm are determined through multiple experiments based on the rule optimization effect. This fusion algorithm can quickly and effectively optimize the detection rules of sensor nodes and improve the overall performance of the inventory monitoring system.

[0030] Furthermore, when storing historical data, the database unit employs a spatiotemporal index-optimized storage algorithm. For data collected by each sensor node, indexes are established according to both time and spatial dimensions. The time index uses timestamps as keys, and the spatial index is encoded based on the physical location of the nodes. The data storage formula is as follows:

[0031] D = I t (T)×I s (L)+D data

[0032] Where D represents the stored data object, and I t (T) is a time indexing function that quickly locates the storage location of data based on the timestamp T. s (L) is a spatial indexing function that determines the region to which the data belongs based on the node location code L. data Based on the actual collected data, this algorithm can significantly improve the efficiency of querying and retrieving historical data, making it easier for the central processing unit to quickly retrieve data for analysis and mining, while reducing data storage redundancy and improving storage resource utilization.

[0033] Furthermore, the human-computer interaction interface employs a visual intelligent recommendation method. When managers log into the system to view data, the system comprehensively generates a data visualization recommendation scheme based on the managers' historical operating habits, current inventory status, and abnormal situations. Specifically, the system considers the frequency of managers' past use of different types of data visualization displays to determine the weight of operating habits; it determines the importance weight of current data based on factors such as the correlation between data and current abnormal situations and the magnitude of data changes; simultaneously, based on the display effects of different visualization methods on different types of data, it determines the adaptability of each data type to the visualization method through experiments and machine learning training. These weights and adaptability are comprehensively calculated to provide managers with personalized and efficient data visualization displays, helping them quickly obtain key information and improve management efficiency.

[0034] Compared with existing technologies, this inventory monitoring system for precious metals has the following advantages:

[0035] First, this system deploys multiple IoT sensor nodes and employs a dynamic threshold adaptive algorithm to perform preliminary analysis and judgment on the collected data. It can dynamically adjust the threshold according to environmental changes, thereby improving the accuracy of anomaly detection and avoiding misjudgments or omissions caused by gradual environmental changes. At the same time, when a node detects abnormal data, the swarm intelligence collaborative network module sends signals and data to surrounding nodes for collaborative verification. It uses a credibility-weighted voting algorithm to comprehensively consider the credibility of each node, improving the reliability of collaborative verification. In addition, when sensor nodes send abnormal signals, a dynamic routing optimization algorithm is used to select the transmission path to ensure that the abnormal signals can be transmitted quickly and stably, further improving the timeliness of anomaly detection.

[0036] Second, when the system further analyzes the abnormal data after collaborative verification through the central processing unit, it adopts an anomaly severity assessment algorithm to determine the weight and quantification value of different anomaly factors based on the impact of the anomaly type on precious metal materials, generates different levels of early warning information, and adopts different notification methods and processing procedures to achieve precise hierarchical management of anomalies. At the same time, the system uses reinforcement learning genetic algorithm fusion optimization algorithm to optimize the detection rules of sensor nodes, improve the overall performance of the inventory monitoring system, and the database unit adopts spatiotemporal index optimization storage algorithm when storing historical data, which greatly improves the query and retrieval efficiency of historical data, making it easier for the central processing unit to quickly retrieve data for analysis and mining, and improving data processing efficiency.

[0037] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0039] Figure 1 This is a flowchart illustrating the functional implementation of an inventory monitoring system for precious metals (gold).

[0040] Figure 2 This is an overall flowchart of an inventory monitoring system for precious metals (gold).

[0041] Figure 3This is a flowchart framework for a key module of an inventory monitoring system for precious metals. Detailed Implementation

[0042] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0043] Example 1

[0044] During the system deployment phase, staff deployed sensor node units in different areas of the precious metal warehouse (such as around the shelves, warehouse entrance, and temperature and humidity sensitive areas). These nodes include various IoT sensors such as temperature and humidity sensors and image sensors. Each node has a built-in microprocessor, communication module, and storage module, which can independently collect environmental data (such as temperature and humidity) and image information in the warehouse and perform preliminary analysis and judgment. This deployment step enables full-area monitoring of the warehouse without blind spots, ensuring that all kinds of data of the gold storage environment can be captured in a timely manner, providing comprehensive raw data support for subsequent anomaly judgment, and spatially covering all areas that may affect the safety of the gold.

[0045] After the sensor nodes start working, they continuously collect data from the area they are in. For environmental data such as temperature and humidity, each node uses a dynamic threshold adaptive algorithm to perform preliminary analysis on the collected data. The algorithm formula is as follows:

[0046]

[0047] Where T new For the updated threshold, T old The original threshold is α, the adaptive adjustment coefficient is x. i For the data collected in the i-th time, The threshold is the average of the most recent n collected data, where n is the data statistical period. The threshold is continuously updated based on historical collected data to identify abnormal environmental data that exceeds the normal range. The application of this algorithm allows the threshold to be automatically adjusted with the natural changes in the warehouse environment (such as the temperature difference between day and night and the temperature and humidity fluctuations caused by seasonal changes), avoiding too many false alarms or missed alarms due to fixed thresholds, ensuring that the judgment of environmental anomalies is more in line with the actual situation, and improving the accuracy of preliminary analysis.

[0048] After acquiring images of the warehouse, the image sensor analyzes them using an anomaly detection method based on feature pyramids and an attention mechanism. First, multi-scale features of the image are extracted through a feature pyramid network to obtain feature maps at different levels. Then, an attention mechanism is introduced to assign corresponding weights to each feature map, and the resulting fusion forms the final feature representation. Subsequently, a classifier and a regressor are used to determine whether there are any anomalous targets in the image (such as unauthorized personnel, abnormally moving objects, etc.) and to determine the location and category of the targets. This method can accurately identify anomalous targets of different sizes and locations. The addition of the attention mechanism allows the system to pay more attention to the features of key areas, such as the area near the gold storage shelves, thereby improving the detection accuracy of anomalous targets and promptly detecting personnel or object activities that may threaten the safety of the gold.

[0049] When a sensor node (node ​​A) detects abnormal data (such as a sudden and significant increase in temperature or the appearance of an abnormal target in an image), it sends a verification request to multiple surrounding nodes through the swarm intelligence collaborative network module. This step can use the detection data of surrounding nodes to confirm the initial anomaly judgment, avoid erroneous detection results caused by a single node due to failure or interference, and enhance the reliability of anomaly judgment.

[0050] Upon receiving the request, surrounding nodes assess the data they have collected and report the result (0 indicating no anomaly, 1 indicating an anomaly) to node A. The swarm intelligence collaborative network module processes these results using a credibility-weighted voting algorithm. When node A detects an anomaly, it notifies surrounding nodes B1, B2, ..., B... m After sending the verification request, each node provides a judgment result based on its own detection data (0 indicates no anomaly, 1 indicates anomaly). The final anomaly judgment result R is determined by the following formula:

[0051]

[0052] Where w j For node B j The credibility weight, r j For node B j The judgment result is m, which is the number of surrounding nodes participating in collaborative verification, and β is the voting threshold. Finally, it is determined whether it is an abnormal situation. This algorithm assigns weights to the credibility of different nodes, allowing nodes with stable long-term performance and accurate detection to play a greater role in voting, thereby improving the effectiveness of collaborative verification and reducing overall errors caused by misjudgments by individual nodes.

[0053] If an anomaly is identified, the swarm intelligence collaborative network module selects the optimal transmission path using a dynamic routing optimization algorithm. This algorithm is an improvement on the ant colony algorithm, and the formula is as follows:

[0054] τ ij(t+1)=(1-ρ)τ ij (t)+Δτ ij (t)

[0055] Where τ ij (t+1) represents the pheromone concentration on the path from node i to node j at time t, ρ is the pheromone evaporation coefficient, and Δτ ij (t) represents the increment of pheromone along the path from node i to node j at time t, Δτ ij (t) represents the pheromone increment on the path from node i to node j at time t. The pheromone increment is calculated using the following formula:

[0056]

[0057] Where Q represents the total amount of pheromones released, and L... k Let be the length of the path traversed by the k-th ant. The abnormal data is transmitted to the central processing unit. This path selection method ensures that the abnormal data is transmitted to the central processing unit at the fastest speed and with the lowest energy consumption, avoiding missing the best time to process the abnormal due to transmission delay, while reducing network congestion and ensuring the efficiency of data transmission.

[0058] After receiving abnormal data, the central processing unit combines historical data and threshold ranges stored in the database unit to analyze the abnormal situation using an anomaly severity assessment algorithm. The algorithm formula is as follows:

[0059]

[0060] Where S represents the severity score of the abnormality, and v k c represents the weight of the k-th anomaly. k Let be the quantitative value of the kth abnormal factor, and s be the number of abnormal factors. The severity score of the abnormality is calculated. This algorithm can quantitatively assess the urgency and scope of impact of abnormal situations, providing a scientific basis for taking different levels of handling measures in the future, and avoiding the waste of resources or inadequate response caused by taking the same handling method for all abnormalities.

[0061] Based on the score range, the central processing unit generates different levels of early warning information and notifies the management personnel through corresponding notification methods. At the same time, the corresponding processing procedures are initiated. The tiered early warning and processing enable the management personnel to quickly understand the severity of the anomaly, prioritize the handling of urgent and serious anomalies, improve response efficiency, and minimize the threat of anomalies to the safety of gold materials.

[0062] The database unit uses a spatiotemporal index-optimized storage algorithm to store relevant data (including collection time, node location, and anomaly content) for this anomaly event, as shown in the formula:

[0063] D = It (T)×I s (L)+D data

[0064] Where D represents the stored data object, and I t (T) is a time indexing function that quickly locates the storage location of data based on the timestamp T. s (L) is a spatial indexing function that determines the region to which the data belongs based on the node location code L. data This storage method, which stores actual collected data for subsequent querying and analysis, allows managers to quickly retrieve historical data by time or location. It facilitates tracing patterns of anomalies, summarizing experiences, and providing data support for optimizing warehouse management and system settings.

[0065] After logging into the system through the human-computer interaction interface, the system generates data visualization recommendation schemes based on the manager's historical operating habits, current inventory status, and abnormal situations using a visual intelligent recommendation method. This allows managers to view monitoring data, early warning information, and historical records. Managers can also use this interface to set system parameters and configure functions. The intelligent recommendation visualization schemes enable managers to obtain key information more intuitively and efficiently, reducing the time cost of viewing data. At the same time, it allows managers to adjust the system according to actual needs, improving the convenience of management.

[0066] The central processing unit also periodically uses reinforcement learning and genetic algorithm fusion optimization algorithms to optimize the detection rules of sensor nodes. Through reinforcement learning, the nodes adjust the detection rules in a simulated environment, and then combine them with genetic algorithm for evolutionary optimization to improve the accuracy and efficiency of detection. The continuously optimized detection rules enable the system to adapt to long-term changes in the warehouse environment, continuously improve the ability to detect anomalies, and ensure the reliability and effectiveness of gold inventory monitoring.

[0067] Example 2

[0068] Sensor node units are deployed in the temporary storage compartments of precious metal transport vehicles. These nodes include vibration sensors, smoke sensors, and image sensors, and also have built-in microprocessors, communication modules, and storage modules. They are used to collect environmental data (such as vibration amplitude and the presence of smoke) and image information in the temporary storage compartments during transportation and to perform preliminary analysis. This deployment enables comprehensive real-time monitoring of the temporary storage compartments during transportation, covering potential safety risks that gold may face while in motion (such as vibration, fire, and illegal intrusion). This builds the first line of defense for subsequent anomaly identification and handling, ensuring that environmental and status data throughout the transportation process are traceable and analyzable.

[0069] When the sensor nodes are working, the vibration sensor and smoke sensor continuously collect data. Each node adopts a dynamic threshold adaptive algorithm, which continuously updates the threshold based on historical data during transportation to determine whether the currently collected data is abnormal (such as excessive vibration amplitude or the presence of smoke). The application of this algorithm allows the threshold to adapt to the complex and ever-changing environment during transportation (such as the difference in the degree of bumps in different road sections, the impact of external temperature changes on the cabin environment, etc.), avoiding the misjudgment of normal transportation vibrations as abnormal due to fixed thresholds, or the failure to detect truly dangerous severe vibrations. This improves the adaptability and accuracy of preliminary data analysis and reduces the interference of invalid warnings on escort work.

[0070] After the image sensor acquires images of the temporary storage compartment, an abnormal target detection method based on feature pyramids and attention mechanisms is used. First, multi-scale feature maps are extracted through a feature pyramid network, and then an attention mechanism is introduced to assign weights and fuse them. Then, a classifier and regressor are used to determine whether there are abnormal targets (such as abnormally opened doors or foreign objects entering the compartment). This method can accurately capture various abnormal targets that may occur during transportation. The attention mechanism allows the system to focus on the image features of key areas such as the gold storage location and doors. Even when the image is slightly shaken due to vehicle movement, it can effectively identify anomalies and promptly detect behaviors that may threaten the safety of the gold (such as illegally opening doors), providing visual monitoring support for transportation safety.

[0071] When a sensor node (node ​​A) detects an anomaly (such as severe vibration or an abnormally opened hatch shown in the image), it sends a verification request to surrounding nodes through the swarm intelligence collaborative network module. This step, with the help of collaborative verification by surrounding nodes, can effectively eliminate false alarms caused by factors such as transportation bumps and signal interference from a single node. By cross-verifying data from multiple nodes, the credibility of anomaly judgment is improved, ensuring that only real anomalies will enter the subsequent processing flow, thus avoiding unnecessary emergency responses.

[0072] Peripheral nodes make judgments based on their own data and report the results. The swarm intelligence collaborative network module uses a credibility-weighted voting algorithm to process these results and determine whether they are real anomalies. This algorithm assigns credibility weights to the historical performance of different nodes (e.g., nodes that have been running stably for a long time and have accurate detection have higher weights), making the voting results more valuable and reducing the impact of individual node failures or misjudgments on the overall judgment. This further improves the reliability of collaborative verification and provides a more accurate basis for subsequent decisions.

[0073] If an anomaly is detected, the swarm intelligence collaborative network module selects the optimal path using a dynamic routing optimization algorithm (an improvement on the ant colony algorithm) to transmit the abnormal data to the central processing unit (which can be deployed in the central control system of the transport vehicle or a remote monitoring center). This path optimization method can quickly find the most stable and efficient data transmission path when the network signal is unstable due to vehicle movement, ensuring that the abnormal information is delivered to the processing center in a timely manner, thus buying time for a rapid response.

[0074] After receiving the data, the central processing unit combines it with historical data and other information stored in the database unit during transportation, and uses an anomaly severity assessment algorithm to analyze the anomaly and calculate a severity score. This algorithm can quantitatively assess the urgency of the anomaly (e.g., excessively high smoke concentration may indicate a fire and requires immediate handling; slight vibration may only need to be recorded) and potential risks, providing a scientific standard for subsequent graded handling, avoiding a "one-size-fits-all" approach to all anomalies, and improving the efficiency of emergency resource utilization.

[0075] Based on the scoring, the central processing unit generates corresponding level of early warning information, notifies relevant personnel through appropriate means (such as sending alarms to the terminal devices of the escort personnel), and initiates corresponding processing procedures (such as prompting the escort personnel to check the storage compartment). The graded early warning allows the escort personnel and the remote monitoring center to quickly distinguish the severity of anomalies, prioritize the handling of high-risk anomalies, and ensure that timely measures can be taken to control risks during transportation, maximizing the safety of gold materials.

[0076] The database unit uses a spatiotemporal index to optimize the storage algorithm and store abnormal data (including collection time, node location, and abnormality type) for easy tracing later. This storage method can accurately associate abnormal data in the transportation process with time (such as a certain road segment or time point) and space (such as a certain location in the storage compartment), which is convenient for reviewing the causes of abnormalities and analyzing risk points in the transportation route. It provides data support for optimizing transportation plans and strengthening the management of weak links.

[0077] Remote administrators log in to the system through a human-computer interaction interface. Based on a visualization and intelligent recommendation method, the system generates visualized data schemes according to the administrators' operating habits, current transportation and storage status, and abnormal situations. This allows administrators to view relevant data and early warning information. Administrators can also adjust system parameters through the interface. The intelligently recommended visualization schemes enable remote administrators to quickly focus on key information (such as the current abnormal location and historical records of similar abnormal handling) in complex transportation data, improving the efficiency of remote monitoring and decision-making. At the same time, it is convenient to adjust system parameters in real time according to changes in the transportation environment, ensuring that monitoring always meets actual needs.

[0078] The central processing unit periodically optimizes the detection rules of sensor nodes using reinforcement learning genetic algorithms and fusion optimization algorithms to adapt to the complex and ever-changing environment during transportation, improve the reliability of anomaly detection, and enable sensor nodes to better adapt to different transportation scenarios (such as vibration differences between urban roads and highways, and environmental changes under different climatic conditions), reduce false alarms or missed alarms caused by environmental changes, and ensure the stability and effectiveness of the monitoring system during long-term transportation.

[0079] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An inventory monitoring system for precious metal gold, characterized in that, The system comprises the following components: sensor node units, swarm intelligence collaborative network modules, central processing units, database units, and human-machine interface; The sensor node unit: Multiple Internet of Things sensor nodes are deployed in the precious metal storage area. Each node has a built-in microprocessor, communication module and storage module, which are used to collect data on the gold storage environment and perform preliminary analysis and judgment. The swarm intelligence collaborative network module connects each sensor node through wireless communication technology. When a node detects abnormal data, it sends signals and data to surrounding nodes, and the surrounding nodes perform collaborative verification. The central processing unit communicates with the sensor node unit and the swarm intelligence collaborative network module, receives, stores and analyzes data, processes abnormal data after collaborative verification and generates early warning information, and manages and configures the network. The database unit stores historical data, threshold ranges, abnormal event records, and node configuration information collected by the sensor nodes. The human-computer interaction interface allows managers to view monitoring data, early warning information, and historical records, and to set system parameters and configure functions.

2. The inventory monitoring system for precious metal gold materials according to claim 1, characterized in that, In the sensor node unit, each sensor node uses a dynamic threshold adaptive algorithm to perform preliminary data analysis and judgment. The algorithm formula is as follows: Where T new For the updated threshold, T old The original threshold is α, the adaptive adjustment coefficient is x. i For the data collected in the i-th time, It is the average value of the most recent n collected data, where n is the data statistical period.

3. The inventory monitoring system for precious metal gold materials according to claim 1, characterized in that, When analyzing images, the image sensor in the sensor node unit employs an anomaly target detection method based on feature pyramids and an attention mechanism. This method first extracts multi-scale features from the image through a feature pyramid network to obtain feature maps at different levels. Then, an attention mechanism is introduced to assign corresponding weights to each feature map. The weighted feature maps are then fused to form the final feature representation. Finally, a classifier and a regressor are used to analyze the final features to determine whether there are any anomaly targets in the image and to determine the location and category of the targets. The parameters of the attention mechanism are obtained by training with a large amount of labeled gold storage scene image data. The classifier and regressor are trained using a combination of cross-entropy loss function and mean squared error loss function.

4. The inventory monitoring system for precious metal gold materials according to claim 1, characterized in that, In the aforementioned swarm intelligence collaborative network module, the collaborative verification between nodes adopts a credibility-weighted voting algorithm. When node A detects an anomaly, it sends a notification to its surrounding nodes B1, B2, ..., B1. m After sending the verification request, each node provides its judgment result based on its own detection data. The final anomaly judgment result R is determined by the following formula: Where w j For node B j The credibility weight, r j For node B j The judgment result is given by m, which is the number of surrounding nodes participating in collaborative verification, and β, which is the voting threshold.

5. The inventory monitoring system for precious metal gold materials according to claim 1, characterized in that, In the aforementioned swarm intelligence collaborative network module, when a sensor node sends an abnormal signal, a dynamic routing optimization algorithm is used to select the transmission path. This algorithm is an improvement on the ant colony algorithm, and the formula is as follows: t ij (t+1)=(1-ρ)τ ij (t)+Δτ ij (t) Where τ ij (t+1) represents the pheromone concentration on the path from node i to node j at time t, where i is the pheromone evaporation coefficient, and Δτ ij (t) represents the increment of pheromone along the path from node i to node j at time t, Δτ ij (t) represents the pheromone increment on the path from node i to node j at time t. The pheromone increment is calculated using the following formula: Where Q represents the total amount of pheromones released, and L... k Let be the length of the path traversed by the k-th ant.

6. The inventory monitoring system for precious metal gold materials according to claim 1, characterized in that, When the central processing unit further analyzes the abnormal data, it employs an anomaly severity assessment algorithm, the formula of which is: Where S represents the severity score of the abnormality, and v k c represents the weight of the k-th anomaly. k S represents the quantification value of the kth abnormal factor, and s represents the number of abnormal factors. Based on the calculated S value, the central processing unit generates different levels of early warning information according to different score ranges and adopts different notification methods and processing procedures.

7. The inventory monitoring system for precious metal gold materials according to claim 1, characterized in that, When optimizing the detection rules of sensor nodes, the central processing unit employs a reinforcement learning-genetic algorithm fusion optimization algorithm. First, reinforcement learning is used to allow sensor nodes to perform data detection and decision-making in different simulated environments, receiving rewards or penalties based on the decision results, and continuously adjusting the detection rules to maximize long-term rewards. Then, a genetic algorithm is introduced to evolve and optimize the rules obtained from reinforcement learning through selection, crossover, and mutation operations, using the following formula: R new =GA(RL(R old )) Where R new For the optimized detection rules, R old The initial detection rules are defined by RL (Reinforcement Learning) and GA (Genetic Algorithm). In the algorithm, the reward function of reinforcement learning is set according to the anomaly detection accuracy and false alarm rate indicators, and the crossover probability and mutation probability of the genetic algorithm are determined through multiple experiments based on the rule optimization effect.

8. The inventory monitoring system for precious metal gold materials according to claim 1, characterized in that, When storing historical data, the database unit employs a spatiotemporal index-optimized storage algorithm. For data collected by each sensor node, indexes are built according to both time and spatial dimensions. The time index uses timestamps as keys, and the spatial index is encoded based on the physical location of the nodes. The data storage formula is as follows: D=I t (T)×I s (L)+D data Where D represents the stored data object, and I t (T) is a time indexing function that quickly locates the storage location of data based on the timestamp T. s (L) is a spatial indexing function that determines the region to which the data belongs based on the node location code L. data This refers to the actual data collected.

9. The inventory monitoring system for precious metal gold materials according to claim 1, characterized in that, The human-computer interaction interface adopts a visual intelligent recommendation method. When managers log in to the system to view data, the system will generate a data visualization recommendation scheme based on the managers' historical operating habits, current inventory status, and abnormal situations. Specifically, the system will consider the frequency of managers' previous use of different types of data visualization to determine the weight of operating habits; determine the importance weight of the current data based on factors such as the correlation between the data and the current abnormal situation and the magnitude of data changes; and determine the suitability of each data type and visualization method based on the display effects of different visualization methods on different types of data through experiments and machine learning training.

Images