Intelligent robot knowledge base management method and system

Abstract

The invention relates to the technical field of intelligent robot knowledge management, in particular to an intelligent robot knowledge base management method and system, and the method comprises the steps: inputting an original data flow generated by robot operation into a knowledge deposition module for deposition and layering, and forming an initial knowledge deposition layer with a hierarchical structure; the initial knowledge deposition layer is imported into a knowledge smelting engine, and the engine carries out cross-layer fusion and recrystallization treatment on the initial knowledge deposition layer to generate a standardized knowledge crystal set. And conveying the knowledge crystal set to a topology braiding device, and performing spatial relation mapping and link braiding on the knowledge crystals through the braiding device to construct a dynamic knowledge topology network. And performing knowledge state evaluation and evolution tracking based on the dynamic knowledge topology network. According to the method, the cognitive depth and the dynamic evolution ability of the knowledge base are improved by forming the structured knowledge units and constructing the adaptive relation network.

Description

Technical Field

[0001] This invention relates to the field of intelligent robot knowledge management technology, and in particular to an intelligent robot knowledge base management method and system. Background Technology

[0002] Current knowledge base construction and management for intelligent robots generally rely on rule-based systems, ontology engineering, or static knowledge graph technologies. These methods typically transform collected data or experience into discrete facts, rules, or entity-relationship triples for storage and retrieval. This type of knowledge representation is flat, lacking an inherent logical structure and hierarchical organization between knowledge units. This makes it difficult for the knowledge base to effectively reflect and express the robot's progressive learning process from raw data to higher-order cognition. Knowledge accumulation is more like a simple stacking of information than organic growth, thus limiting the robot's deep reasoning and adaptive learning capabilities.

[0003] Existing technical solutions have shortcomings in handling knowledge updates and system evolution. Once a static knowledge graph or rule base is built, its structure is relatively fixed. When new knowledge needs to be incorporated or new relationships between knowledge need to be reflected, it often requires manual local corrections or complex global reconstruction, a cumbersome process with a delayed response. The knowledge system lacks an inherent, proactive adjustment mechanism; the links are rigid, and it cannot dynamically reorganize and expand in real time based on changes in the knowledge content itself and the context, making it difficult for the knowledge base to adapt to complex and unstable real-world operating environments.

[0004] A key issue is how to effectively transform the massive, heterogeneous raw data generated during robot operation into advanced knowledge units with stable internal structures and the ability to be freely combined. Simultaneously, another challenge is how to enable these knowledge units to autonomously and dynamically organize into a continuously evolving relational network, rather than a static graph, to truly reflect the process of knowledge systems growing and being reconstructed with accumulated experience. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an intelligent robot knowledge base management method and system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for managing an intelligent robot knowledge base, comprising: The raw operational data stream generated by the intelligent robot during operation is input into the knowledge deposition module; The knowledge deposition module deposits and layers the original running data stream to form an initial knowledge deposition layer with a hierarchical structure. The initial knowledge deposition layer is imported into the knowledge melting engine, which then performs cross-layer fusion and recrystallization on the initial knowledge deposition layer to generate a standardized set of knowledge crystals. The knowledge crystal set is fed to the topology weaver; A dynamic knowledge topology network is constructed by mapping and linking the knowledge crystals in the knowledge crystal set using a topology weaver. The evaluation and evolution tracking of knowledge state are performed based on dynamic knowledge topology networks.

[0007] As a further aspect of the present invention, the deposition and layering of the original running data stream by the knowledge deposition module to form an initial knowledge deposition layer with a hierarchical structure specifically includes: The knowledge deposition module receives the original running data stream and performs timeliness marking and source classification on the original running data stream; Density analysis and event slicing are performed on the raw running data stream that has been classified to extract key event segments from the data stream; Based on the semantic density and temporal continuity of key event fragments, different key event fragments are deposited into different data layers; Perform intra-layer calibration and timestamp alignment on key event fragments deposited into each data layer; Integrate all calibrated and aligned data levels to output the initial knowledge deposition layer with a hierarchical structure.

[0008] As a further aspect of the present invention, the knowledge melting engine performs cross-layer fusion and recrystallization on the initial knowledge deposition layer to generate a standardized set of knowledge crystals, specifically including: The knowledge melting engine parses the hierarchical structure of the initial knowledge deposition layer and reads key event fragments within each layer; Search for key event fragments with semantic or causal relationships between different levels of the initial knowledge deposition layer; The identified key event fragments are then integrated and strung together with context to form preliminary knowledge blocks. The initial knowledge blocks are structurally regularized and their boundaries are defined to eliminate contradictions and redundant information within the knowledge blocks; Encapsulate the completed and defined knowledge blocks into knowledge crystals with a unified interface format; All the packaged knowledge crystals are gathered together to form the standardized knowledge crystal set.

[0009] As a further aspect of the present invention, constructing a dynamic knowledge topology network by mapping and linking the knowledge crystals in the knowledge crystal set using a topology weaver specifically includes: The topology weaver acquires the set of knowledge crystals and parses the core semantic tags and attribute fields of each knowledge crystal; Between any two knowledge crystals, the association weight between the knowledge crystals is calculated based on the similarity of the core semantic tags and the complementarity of the attribute fields; Based on the pairwise association weights between all knowledge crystals, an initial relationship graph is constructed with knowledge crystals as nodes and association weights as edge weights; A time dimension constraint is introduced into the initial relation graph, and dynamic decay and enhancement operations are performed on the edge weights to make the relation graph have temporal dynamics. The relationship graph with temporal dynamics is optimized by path optimization and redundant edge pruning, and finally the dynamic knowledge topology network is output.

[0010] As a further aspect of the present invention, the calculation of the association weight between knowledge crystals based on the similarity of core semantic tags and the complementarity of attribute fields specifically includes: Extract the core semantic tag sets of the first and second knowledge crystals; Calculate the Jaccard similarity between the core semantic label set of the first knowledge crystal and the core semantic label set of the second knowledge crystal, and use it as the semantic similarity. Extract the attribute field sets of the first knowledge crystal and the second knowledge crystal respectively; Identify the field combinations in the attribute field set of the first knowledge crystal that can form a complete information chain with the attribute field set of the second knowledge crystal; The number of field combinations that can form a complete information chain is counted, and the ratio of the number of field combinations to the total number of fields in the two knowledge crystal attribute field sets is used as the attribute complementarity. The semantic similarity and attribute complementarity are weighted and summed, and the result is normalized to the range of zero to one, which is used as the association weight between the first knowledge crystal and the second knowledge crystal.

[0011] As a further aspect of the present invention, the step of introducing a time dimension constraint into the initial relation graph and performing dynamic decay and enhancement operations on the edge weights to give the relation graph temporal dynamics specifically includes: Add a timestamp to each edge in the initial relation graph. The timestamp records the last time that the two knowledge crystals connected by the edge became associated. Get the time difference between the current system time and the timestamp corresponding to each edge; Define an attenuation coefficient function, the input of which is the time difference, and the output is an attenuation factor that decreases as the time difference increases; For each edge, its original association weight is multiplied by the corresponding decay factor to obtain the base weight after time decay; Monitor new interaction events between knowledge crystals, and when a new interaction event is detected, obtain the knowledge crystal pair involved in the event; Find the edge corresponding to the knowledge crystal pair, and increase the weight of the corresponding edge by an enhancement amount based on the event intensity; The weights of all edges after decay and enhancement operations are normalized as a whole, and the initial relation graph is updated to give the relation graph temporal dynamics.

[0012] As a further aspect of the present invention, the evaluation and evolution tracking of knowledge state based on a dynamic knowledge topology network specifically includes: Establish a state probe network connected to the dynamic knowledge topology network. The state probe network is responsible for periodically collecting the activity index of nodes and the flow intensity of edges in the dynamic knowledge topology network. The activity indicators and flow intensity collected by the state probe network are input into the evolution analyzer; The evolutionary analyzer fits the overall stability curve and local activity distribution map of the dynamic knowledge topology network based on the historical sequence of activity indicators and flow intensity. The overall stability curve and local activity distribution map are compared with the preset knowledge health benchmark model; Based on the comparison results, rigid nodes and overloaded links in the dynamic knowledge topology network were identified.

[0013] As a further aspect of the present invention, the comparison of the overall stability curve and the local activity distribution map with a preset knowledge health benchmark model specifically includes: Obtain a preset knowledge health benchmark model, which includes a standard overall stability curve template and a standard local activity distribution map template; The overall stability curve obtained by fitting the evolution analyzer is compared point by point with the standard overall stability curve template, and the sum of squares of the differences between the two curves at each time point is calculated as the overall deviation. The local activity distribution map generated by the evolutionary analyzer is compared with the standard local activity distribution map template by dividing the region, and the cosine similarity of the activity values ​​of the corresponding region is calculated as the local similarity. Based on the overall deviation and local similarity, calculate the comprehensive difference index between the current state of the dynamic knowledge topology network and the knowledge health benchmark model; The overall health status of the dynamic knowledge topology network is determined by comparing the comprehensive difference index with a preset threshold.

[0014] As a further aspect of the present invention, after identifying the stagnant nodes and overloaded links in the dynamic knowledge topology network based on the comparison results, the method further includes: A knowledge waking process is initiated for the identified stagnant nodes, which includes injecting new association query requests or association context information into the stagnant nodes. To initiate a traffic diversion process for the identified overloaded links, the traffic diversion process includes building new auxiliary connections near the overloaded links or queuing and scheduling knowledge requests flowing to the overloaded links. The updated status of the stagnant nodes processed by the knowledge awakening process and the overloaded links processed by the traffic diversion process are fed back to the dynamic knowledge topology network to complete the real-time adjustment of the dynamic knowledge topology network.

[0015] As a further aspect of the present invention, the present invention also includes an intelligent robot knowledge base management system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the intelligent robot knowledge base management method described above.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: The knowledge fusion engine performs cross-layer fusion and recrystallization on the initial knowledge deposition layer. Cross-layer fusion breaks down the barriers between data at different levels, enabling deep interaction and semantic alignment between the underlying raw features and the higher-level abstract concepts. The recrystallization process, based on this fusion, standardizes, reconstructs, and encapsulates the information, forming knowledge crystals with clear boundaries and complete structures. This processing transforms knowledge from loose, flat fragments into independent cognitive modules with clear semantics and stable interfaces. Knowledge crystals possess excellent composability and reusability, providing appropriately granular and richly meaningful operational objects for complex reasoning, thus improving the accuracy and structure of knowledge representation.

[0017] The topology weaver maps and weaves spatial relationships within a collection of knowledge crystals. It doesn't merely establish static connections based on preset rules; instead, it actively calculates and dynamically generates the association weights and link types between crystals based on their inherent attributes, current context, and historical interactions. This process constructs a dynamic knowledge topology network whose node links can be adjusted, strengthened, weakened, or reconstructed in real time as new knowledge crystals are integrated, the state of old knowledge changes, and external queries stimulate the system. The overall network structure possesses an organism-like growth and adaptability, enabling the entire knowledge system to continuously evolve, instantly reflecting the latest cognitive states and association patterns, thus transforming the knowledge base from static storage to a dynamic, living system. Attached Figure Description

[0018] Figure 1 This is a flowchart of the intelligent robot knowledge base management method described in this invention; Figure 2A flowchart for constructing a dynamic knowledge topology network; Figure 3 A line graph showing the changes in node activity in a dynamic knowledge topology network; Figure 4 A heatmap of local activity in a knowledge topology network; Figure 5 A multi-dimensional performance radar chart before and after traffic scheduling. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 The system inputs the raw operational data stream generated by the intelligent robot during operation into the knowledge deposition module. The knowledge deposition module deposits and layers the raw operational data stream to form an initial knowledge deposition layer with a hierarchical structure. The initial knowledge deposition layer is then imported into the knowledge melting engine, which performs cross-layer fusion and recrystallization on the initial knowledge deposition layer to generate a standardized set of knowledge crystals. The set of knowledge crystals is then sent to the topology weaver. The topology weaver maps and links the knowledge crystals in the set to spatial relationships, constructing a dynamic knowledge topology network. Based on the dynamic knowledge topology network, the system performs knowledge state evaluation and evolution tracking.

[0022] In one embodiment of the present invention, see [reference] Figure 2The intelligent robot "HomeBot" operates in a home environment. Its raw operational data stream includes image sequences captured by a camera, audio streams collected by a microphone, point cloud data generated by LiDAR, and task execution log text. The knowledge deposition module receives this raw data stream and performs time-sensitivity marking and source classification. Time-sensitivity marking uses timestamps to record the data generation time, and source classification is based on the data sensor type, categorized as visual, auditory, spatial perception, and log-based. Density analysis and event slicing are performed on the classified raw data stream to extract key event fragments. Density analysis calculates the information entropy value of the data stream within a time window to identify density regions. Event slicing extracts continuous data segments as key event fragments based on the boundaries of density regions. In some embodiments, density analysis uses a sliding window method with an adjustable window size to accommodate different data rates. Based on the semantic density and temporal continuity of the key event fragments, different key event fragments are deposited into different data levels. Semantic density is quantified by analyzing the frequency of text keywords or the richness of image features within the fragment, while temporal continuity checks whether the time interval between fragments is less than a threshold. The key event fragments deposited into each data level are calibrated and timestamp aligned within the layer. The calibration operation unifies the data format and encoding of the fragments, and the alignment operation ensures that the timestamps of the fragments within the layer are monotonically increasing and conflict-free. All calibrated and aligned data levels are integrated to output an initial knowledge deposition layer with a hierarchical structure. The initial knowledge deposition layer includes a base layer, an intermediate layer, and a high layer. The base layer stores the original perception fragments, the intermediate layer stores the abstract event descriptions, and the high layer stores the task sequences.

[0023] In practical implementation, the knowledge fusion engine parses the hierarchical structure of the initial knowledge deposition layer and reads key event fragments within each layer. The parsing operation traverses the hierarchical tree and loads fragment metadata. It searches for key event fragments with semantic or causal relationships between different layers of the initial knowledge deposition layer. Semantic relationships are determined by comparing the co-occurrence of fragment tags, while causal relationships are identified based on temporal sequence and logical reasoning. The found related key event fragments are then fused and contextually spliced ​​to form preliminary knowledge blocks. The fusion operation merges fragment content, and the splicing operation connects fragment contexts to form a coherent narrative. In some embodiments, content fusion uses natural language processing technology to integrate text descriptions, while contextual splicing preserves temporal order and logical relationships. The preliminary knowledge blocks undergo structural regularization and boundary definition to eliminate contradictions and redundant information within the knowledge blocks. Structural regularization reorganizes the arrangement of knowledge elements, and boundary definition clarifies the subject scope of the knowledge blocks. It is understood that contradiction detection is achieved through logical consistency checks, and redundancy elimination removes duplicate or highly similar content. The standardized and defined knowledge blocks are encapsulated into knowledge crystals with a unified interface format. The encapsulation process adds standardized metadata such as unique identifiers, creation time, and version numbers. Optionally, the unified interface format uses a JSON-LD structure to support semantic interoperability. All encapsulated knowledge crystals are then aggregated to form a standardized knowledge crystal set. This set is stored in a distributed database for subsequent module access. The standardization of the knowledge crystal set is reflected in each knowledge crystal adhering to the same pattern and access protocol.

[0024] In one embodiment of the invention, the intelligent robot "NavBot" operates in a warehouse environment. Its generated standardized knowledge crystal set includes knowledge crystals describing "identifying shelf A," "moving to area B," "low battery alarm," and "planning charging path." A topology weaver acquires the knowledge crystal set and parses the core semantic tags and attribute fields of each knowledge crystal. The parsing operation reads the metadata of the knowledge crystal to extract the core semantic tag set and attribute field set. Between any two knowledge crystals, the association weight between them is calculated based on the similarity of the core semantic tags and the complementarity of the attribute fields. The calculation process involves extracting and comparing the tag sets and attribute sets of the two knowledge crystals. In some embodiments, the tag similarity calculation uses the Jaccard index, and the attribute complementarity judgment is based on whether the field values ​​can form a logical closed loop. An initial relationship graph is constructed based on the pairwise association weights of all knowledge crystals, with knowledge crystals as nodes and association weights as edge weights. The construction operation maps each knowledge crystal to a graph node and the association weight to the weight value of the edge connecting the corresponding node. A time-dimensional constraint is introduced into the initial relational graph to dynamically decay and enhance edge weights, giving the graph temporal dynamism. This time constraint is manifested in each edge being associated with a last updated timestamp. Decay operations reduce weights based on the difference between the current time and the timestamp, while enhancement operations increase weights in response to new interaction events. This dynamism allows the relational graph to reflect changes in the popularity of knowledge connections. The temporally dynamic relational graph is then subjected to path optimization and redundant edge pruning to ultimately output a dynamic knowledge topology network. Path optimization merges multiple indirect connections into direct connections, and redundant edge pruning removes edges with weights below a set threshold. Optionally, redundant edge pruning employs a community detection-based algorithm to preserve core connections within communities.

[0025] In the specific implementation, specific operations are performed when calculating the association weight between knowledge crystals. For knowledge crystal A, its core semantic label set is {"navigation", "shelf", "recognition"} and its attribute field set is {"coordinate X:100", "coordinate Y:200", "camera ID:1"}. For knowledge crystal B, its core semantic label set is {"navigation", "transportation", "B area"} and its attribute field set is {"starting coordinate X:100", "starting coordinate Y:200", "target area: B"}. The Jaccard similarity between the core semantic label sets is calculated as the semantic similarity. The intersection of the labels of knowledge crystal A and knowledge crystal B is {"navigation"}, and the union of the labels is {"navigation", "shelf", "recognition", "transportation", "B area"}. The semantic similarity value is 1 / 5, or 0.2. The attribute field set of knowledge crystal A is identified as having a complete information chain with the attribute field set of knowledge crystal B. For example, the fields "coordinate X:100" and "coordinate Y:200" of knowledge crystal A and "starting coordinate X:100" and "starting coordinate Y:200" of knowledge crystal B numerically match, forming a continuity of positional information. This combination is identified as forming a complete information chain, and the number of field combinations forming a complete information chain is counted as 1. The total number of fields in the two knowledge crystal attribute field sets is 7, and the attribute complementarity value is 1 / 7, approximately 0.143. The weighted sum of semantic similarity and attribute complementarity is normalized to the range of zero to one and used as the association weight between knowledge crystal A and knowledge crystal B. The weighted sum is calculated using the formula:

[0026] in: Indicates the association weight. Indicates semantic similarity. Indicates the complementarity of attributes. It is a preset balance coefficient with a value between 0 and 1. Substitute the values ​​into the calculation The normalization operation linearly maps the calculation result to the [0,1] interval. In this example, the calculated value is already within the interval, so the association weight is determined to be 0.183. In some embodiments, the weighting coefficient λ can be configured according to the characteristics of the knowledge domain.

[0027] In one embodiment of the present invention, the intelligent robot "NavBot" operates in a warehouse environment. The knowledge crystal set includes knowledge crystal K1 describing "identifying shelf A", knowledge crystal K2 describing "moving to area B", and knowledge crystal K3 describing "low battery alarm". The core semantic tag sets of knowledge crystals K1 and K2 are extracted. The core semantic tag set of knowledge crystal K1 is {"identification", "shelf", "location"}, and the core semantic tag set of knowledge crystal K2 is {"moving", "navigation", "area B"}. The Jaccard similarity between the core semantic tag sets of knowledge crystals K1 and K2 is calculated as the semantic similarity. The Jaccard similarity is calculated as the intersection size of the two sets divided by the union size. Attribute field sets are extracted from knowledge crystals K1 and K2 respectively. The attribute field set of knowledge crystal K1 is {"Shelf ID: A001", "Coordinates: (10, 20)"}, and the attribute field set of knowledge crystal K2 is {"Target Coordinates: (30, 40)", "Priority: High"}. Field combinations from the attribute field set of knowledge crystal K1 that can form a complete information chain with the attribute field set of knowledge crystal K2 are identified. These field combinations must satisfy logical coherence, such as the coordinate sequence forming a path planning basis. The number of field combinations that can form a complete information chain is counted. The ratio of the number of field combinations to the total number of fields in the attribute field sets of the two knowledge crystals is used as the attribute complementarity. The weighted sum of semantic similarity and attribute complementarity is normalized to the range of zero to one and used as the association weight between knowledge crystals K1 and K2. A balancing coefficient is used in the weighted sum to adjust the proportion of semantic and attribute contributions. It can be understood that the identification of a complete information chain relies on predefined logical relationship rules in the domain knowledge graph. In some embodiments, Jaccard similarity calculation is implemented using a standard set operation library, and attribute complementarity statistics are performed using an iterative matching algorithm.

[0028] In the specific implementation, the initial relationship graph is constructed based on the association weights. A timestamp is added to each edge in the initial relationship graph. The timestamp records the last time the two knowledge crystals connected by the edge were associated. For example, the timestamp of the edge connecting knowledge crystal K1 and knowledge crystal K2 is recorded as 2023-10-01 10:00:00. The time difference between the current system time and the timestamp corresponding to each edge is obtained. The current system time is 2023-10-04 10:00:00, and the time difference is 3 days. A decay coefficient function is set. The input of the decay coefficient function is the time difference, and the output is a decay factor that decreases as the time difference increases. The decay coefficient function is defined as follows:

[0029] in: Indicates the attenuation factor. The time difference is expressed in days. This is the decay base value, ranging from 0 to 1. For each edge, its original association weight is multiplied by the corresponding decay factor to obtain the base weight after time decay. The original association weight of the edge between knowledge crystal K1 and knowledge crystal K2 is 0.85, and the decay base value is set accordingly. Calculate the attenuation factor The basic weight is .

[0030] The system monitors new interaction events between knowledge crystals. When a new interaction event is detected, the knowledge crystal pair involved in the event is obtained. For example, knowledge crystals K1 and K3 interacted due to a low battery decision on 2023-10-04 11:00:00. The system finds the edge corresponding to knowledge crystals K1 and K3 and adds an enhancement amount based on the event intensity to the weight of the corresponding edge. The event intensity is quantized by the interaction frequency, and the enhancement amount is set to 0.1. The weights of all edges after decay and enhancement operations are then normalized, and the initial relationship graph is updated to give it temporal dynamics. The normalization operation linearly scales all weights to a sum of 1. It can be understood that the decay base value γ can be adjusted according to the application scenario to control the forgetting rate of knowledge associations, and the enhancement amount is dynamically calculated based on the importance of the event. In some embodiments, the decay coefficient function can also adopt a linear decay form to adapt to simple scenarios. Optionally, the time difference unit can be hours or minutes to adapt to high-frequency update environments. Optionally, the enhancement operation is triggered only when the interaction event meets specific logical conditions.

[0031] See Figure 3 This is a line graph showing the changes in node activity of a dynamic knowledge topology network. It illustrates the changes in node activity of three core knowledge crystals (K1, K2, and K3) in the intelligent robot's knowledge base over a 7-day monitoring period, clearly reflecting the dynamic evolution of these knowledge crystals. The high stability of K1 and the suddenness of K3 correspond to "basic core knowledge" and "emergency response knowledge," respectively, providing a quantitative basis for resource allocation and priority scheduling in the knowledge base. The synchronized fluctuations in the activity of K1 and K2, and the abrupt change in the timing of K3, can verify the robot's operational processes and event triggering logic. For example, the peak on day 4 corresponds to a continuous task scenario of "shelf recognition + handling + low battery alarm." The fact that K3's activity remains above 0.70 on day 7 indicates that the subsequent impact of the low battery alarm persists, prompting maintenance personnel to pay attention to the robot's battery health or workload.

[0032] In one embodiment of the present invention, the intelligent robot "InfoBot" operates in an information service center. Its dynamic knowledge topology network contains hundreds of knowledge crystal nodes and thousands of associated edges. A state probe network is established connected to the dynamic knowledge topology network. The state probe network is responsible for periodically collecting the activity indicators of nodes and the flow intensity of edges in the dynamic knowledge topology network. The activity indicators quantify the frequency of node queries or updates, and the flow intensity quantifies the flow of knowledge associations carried by edges that are activated. The activity indicators and flow intensity collected by the state probe network are input into an evolutionary analyzer. The collection period of the state probe network is set to once per minute, and the collected data includes node activity indicator values ​​and edge flow intensity values. The evolutionary analyzer fits the overall stability curve and local activity distribution map of the dynamic knowledge topology network based on the historical sequence of activity indicators and flow intensity. The fitting operation uses time series analysis methods. The overall stability curve reflects the changing trend of the global connection strength of the network, and the local activity distribution map shows a heat map of the activity level of nodes in different regions of the network. The overall stability curve and local activity distribution map are compared with a preset knowledge health benchmark model, which is generated based on data from historical normal operation periods. Based on the comparison results, stagnant nodes and overloaded links in the dynamic knowledge topology network were identified. Stagnant nodes are characterized by an activity index that is consistently below a threshold, while overloaded links are characterized by a flow intensity that is consistently above a threshold. In some embodiments, the state probe network adopts a distributed deployment mode to simultaneously collect data from different parts of the dynamic knowledge topology network, and the evolution analyzer uses a sliding window mechanism to process historical sequence data. In a specific implementation, refer to Table 1, which shows some of the data obtained by the state probe network within one collection cycle.

[0033] Table 1: Partial Data Acquired by the Status Probe Network

[0034] In practical implementation, a pre-defined knowledge health benchmark model is obtained. This model includes a standard overall stability curve template and a standard local activity distribution map template. The standard overall stability curve template is composed of the average stability index at the same time each day during historical health periods. The standard local activity distribution map template divides the network into eight regions and defines the expected activity range for each region. The overall stability curve obtained by the evolutionary analyzer is compared point-by-point with the standard overall stability curve template. The sum of the squares of the differences between the two curves at each time point is calculated as the overall deviation. The overall deviation is calculated using the following formula:

[0035] in: Indicates the overall deviation. Indicates a point-in-time index. Indicates the total number of time points. This indicates the overall stability curve obtained by the evolution analyzer fitting at time point [time value missing]. The value, This indicates the standard overall stability curve template at time point. The value of the local activity distribution map generated by the evolution analyzer is compared with the standard local activity distribution map template by dividing the region. The cosine similarity of the activity values ​​of the corresponding regions is calculated as the local similarity. The activity value of each region is a multi-dimensional vector. The comprehensive difference index between the current state of the dynamic knowledge topology network and the knowledge health benchmark model is calculated based on the overall deviation and local similarity. The comprehensive difference index is an increasing function of the overall deviation and a decreasing function of the local similarity. The comprehensive difference index is compared with a preset threshold to determine the overall health status of the dynamic knowledge topology network. The preset threshold is set according to operation and maintenance experience. It can be understood that the larger the overall deviation, the worse the network stability, and the lower the local similarity, the more the network activity pattern deviates from the normal pattern. In some embodiments, the region division is based on the functional category of the knowledge crystal. Optionally, the comprehensive difference index can be calculated in the form of a weighted harmonic average. Optionally, the standard template will be retrained periodically with new data to adapt to the normal evolution of the dynamic knowledge topology network.

[0036] See Figure 4 This is a heatmap showing the local activity of a knowledge topology network. It displays the local activity distribution of eight knowledge regions in the intelligent robot's knowledge base within a single day (divided into eight 3-hour time slots). Darker colors (purple) indicate lower activity, while brighter colors (yellow) indicate higher activity. Based on the distribution of high-activity periods (6-9h, 12-15h, 18-21h), computing resources can be increased during these periods to improve the knowledge base's response speed. Knowledge updates or model training can be performed during low-activity periods (0-3h, 21-24h) to avoid impacting business operations. The persistently low activity in region 8 may indicate a "stuck node," requiring the initiation of a knowledge wake-up process; the high-frequency activity in region 5 may indicate an "overloaded link," requiring traffic diversion. The time-based distribution of activity closely matches the operational patterns of the warehousing scenario, validating the adaptability of the knowledge base to the business scenario.

[0037] In one embodiment of the present invention, the intelligent robot "TutorBot" operates on an online education platform. Its dynamic knowledge topology network, after evaluation, identifies stagnant nodes and overloaded links. Stagnant nodes are characterized by an activity index for the knowledge crystal "Trigonometric Function Historical Background" consistently below the threshold of 0.2, while overloaded links are characterized by a flow intensity on the edge connecting the knowledge crystals "Proof of the Pythagorean Theorem" and "Application of Hypotenuse Calculation" consistently above the threshold of 0.08. A knowledge awakening process is initiated for the identified stagnant nodes. This process includes injecting new related query requests or related contextual information into the stagnant nodes. For example, the "Trigonometric Function Historical Background" node is injected with the related query request "Please explain the origin of trigonometric functions in conjunction with the development of astronomy and calendar systems," along with the related contextual information "Astronomical observation records recorded on ancient Babylonian clay tablets." To identify overloaded links, a traffic diversion process is initiated. This process includes building new auxiliary connections near the overloaded links or queuing and scheduling knowledge requests flowing to the overloaded links. Auxiliary connections connecting to the "similar triangle property" of the knowledge crystal are built near the "Pythagorean theorem proof" node and the "hypotenuse calculation application" node. Simultaneously, some computational requests flowing to the original overloaded links are queued and prioritized. The updated states of the stagnant nodes processed by the knowledge awakening process and the overloaded links processed by the traffic diversion process are fed back to the dynamic knowledge topology network, enabling real-time adjustments to the network. The activity index values ​​of the update nodes and the flow strength values ​​of the edges are fed back, and the network structure is recalculated.

[0038] In practical implementation, after the knowledge awakening process injects a related query request into the stagnant node "Historical Background of Trigonometric Functions," the node is activated and forms a new association with the knowledge crystal "Ancient Astronomical Instruments," increasing the node activity index from 0.1 to 0.25. The traffic diversion process queues and schedules knowledge requests flowing to the overloaded link, allocating processing order based on the urgency and computational complexity of the request. In some embodiments, the queuing and scheduling employs a multi-level feedback queue mechanism to accommodate requests with different response time requirements. After constructing a new auxiliary connection, the flow intensity of the original overloaded link "Proof of Pythagorean Theorem" - "Application of Hypotenuse Calculation" decreases from 0.15 to 0.06, while the flow intensity of the new auxiliary connection "Proof of Pythagorean Theorem" - "Properties of Similar Triangles" is initialized to 0.05.

[0039] It is understandable that constructing new auxiliary connections essentially adds alternative knowledge-related paths to the dynamic knowledge topology network to distribute query pressure. After the updated state is fed back to the dynamic knowledge topology network, the network recalculates the metrics of nodes and edges, potentially triggering a new round of evaluation and adjustment cycles. Optionally, the content injected into the association query request is dynamically generated by the natural language generation module based on the domains where nodes have missing historical associations. Optionally, the location for constructing auxiliary connections is determined based on potential high-value connection points recommended by the network community discovery algorithm.

[0040] See Figure 5 This is a multi-dimensional performance radar chart showing the performance of the intelligent robot's knowledge base before and after traffic scheduling (red and green). The chart compares five core performance indicators before (red) and after (green) traffic scheduling, visually reflecting the effectiveness of the optimization. The green curve (after scheduling) outperforms the red curve (before scheduling) in all dimensions except "response time," with the largest improvement in "queue length." This indicates that the traffic scheduling (such as building auxiliary connections and request queuing scheduling) successfully resolved the congestion problem of overloaded links, resulting in a comprehensive performance improvement. The radar chart quantifies the effect of traffic diversion, verifying the effectiveness of the "building auxiliary connections + request queuing scheduling" strategy and providing a reusable solution for subsequent dynamic optimization of the knowledge base. The improvement in "resource utilization" indicates more efficient resource allocation after scheduling, serving as a reference for adjusting computing power investment. The optimization of "response time" and "request success rate" directly improves the robot's operational efficiency and task reliability, increasing customer satisfaction.

[0041] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications 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 protection scope of the present invention.

Claims

1. A method for managing a knowledge base for intelligent robots, characterized in that, The method includes: The raw operational data stream generated by the intelligent robot during operation is input into the knowledge deposition module; The knowledge deposition module deposits and layers the original running data stream to form an initial knowledge deposition layer with a hierarchical structure. The initial knowledge deposition layer is imported into the knowledge melting engine, which then performs cross-layer fusion and recrystallization on the initial knowledge deposition layer to generate a standardized set of knowledge crystals. The knowledge crystal set is fed to the topology weaver; A dynamic knowledge topology network is constructed by mapping and linking the knowledge crystals in the knowledge crystal set using a topology weaver. The evaluation and evolution tracking of knowledge state are performed based on dynamic knowledge topology networks.

2. The intelligent robot knowledge base management method as described in claim 1, characterized in that, The knowledge deposition module deposits and layers the original running data stream to form an initial knowledge deposition layer with a hierarchical structure, specifically including: The knowledge deposition module receives the original running data stream and performs timeliness marking and source classification on the original running data stream; Density analysis and event slicing are performed on the raw running data stream that has been classified to extract key event segments from the data stream; Based on the semantic density and temporal continuity of key event fragments, different key event fragments are deposited into different data layers; Perform intra-layer calibration and timestamp alignment on key event fragments deposited into each data layer; Integrate all calibrated and aligned data levels to output the initial knowledge deposition layer with a hierarchical structure.

3. The intelligent robot knowledge base management method as described in claim 1, characterized in that, The knowledge fusion engine performs cross-layer fusion and recrystallization on the initial knowledge deposit layer to generate a standardized set of knowledge crystals, specifically including: The knowledge melting engine parses the hierarchical structure of the initial knowledge deposition layer and reads key event fragments within each layer; Search for key event fragments with semantic or causal relationships between different levels of the initial knowledge deposition layer; The identified key event fragments are then integrated and strung together with context to form preliminary knowledge blocks. The initial knowledge blocks are structurally regularized and their boundaries are defined to eliminate contradictions and redundant information within the knowledge blocks; Encapsulate the completed and defined knowledge blocks into knowledge crystals with a unified interface format; All the packaged knowledge crystals are gathered together to form the standardized knowledge crystal set.

4. The intelligent robot knowledge base management method as described in claim 1, characterized in that, By mapping and linking the knowledge crystals in the knowledge crystal set using a topology weaver, a dynamic knowledge topology network is constructed, specifically including: The topology weaver acquires the set of knowledge crystals and parses the core semantic tags and attribute fields of each knowledge crystal; Between any two knowledge crystals, the association weight between the knowledge crystals is calculated based on the similarity of the core semantic tags and the complementarity of the attribute fields; Based on the pairwise association weights between all knowledge crystals, an initial relationship graph is constructed with knowledge crystals as nodes and association weights as edge weights; A time dimension constraint is introduced into the initial relation graph, and dynamic decay and enhancement operations are performed on the edge weights to make the relation graph have temporal dynamics. The relationship graph with temporal dynamics is optimized by path optimization and redundant edge pruning, and finally the dynamic knowledge topology network is output.

5. The intelligent robot knowledge base management method as described in claim 4, characterized in that, The calculation of association weights between knowledge crystals based on the similarity of core semantic tags and the complementarity of attribute fields specifically includes: Extract the core semantic tag sets of the first and second knowledge crystals; Calculate the Jaccard similarity between the core semantic label set of the first knowledge crystal and the core semantic label set of the second knowledge crystal, and use it as the semantic similarity. Extract the attribute field sets of the first knowledge crystal and the second knowledge crystal respectively; Identify the field combinations in the attribute field set of the first knowledge crystal that can form a complete information chain with the attribute field set of the second knowledge crystal; The number of field combinations that can form a complete information chain is counted, and the ratio of the number of field combinations to the total number of fields in the two knowledge crystal attribute field sets is used as the attribute complementarity. The semantic similarity and attribute complementarity are weighted and summed, and the result is normalized to the range of zero to one, which is used as the association weight between the first knowledge crystal and the second knowledge crystal.

6. The intelligent robot knowledge base management method as described in claim 4, characterized in that, The specific steps of introducing time-dimensional constraints into the initial relation graph and performing dynamic decay and enhancement operations on edge weights to give the relation graph temporal dynamics include: Add a timestamp to each edge in the initial relation graph. The timestamp records the last time that the two knowledge crystals connected by the edge became associated. Get the time difference between the current system time and the timestamp corresponding to each edge; Define an attenuation coefficient function, the input of which is the time difference, and the output is an attenuation factor that decreases as the time difference increases; For each edge, its original association weight is multiplied by the corresponding decay factor to obtain the base weight after time decay; Monitor new interaction events between knowledge crystals, and when a new interaction event is detected, obtain the knowledge crystal pair involved in the event; Find the edge corresponding to the knowledge crystal pair, and increase the weight of the corresponding edge by an enhancement amount based on the event intensity; The weights of all edges after decay and enhancement operations are normalized as a whole, and the initial relation graph is updated to give the relation graph temporal dynamics.

7. The intelligent robot knowledge base management method as described in claim 1, characterized in that, The evaluation and evolution tracking of knowledge state based on dynamic knowledge topology networks specifically includes: Establish a state probe network connected to the dynamic knowledge topology network. The state probe network is responsible for periodically collecting the activity index of nodes and the flow intensity of edges in the dynamic knowledge topology network. The activity indicators and flow intensity collected by the state probe network are input into the evolution analyzer; The evolutionary analyzer fits the overall stability curve and local activity distribution map of the dynamic knowledge topology network based on the historical sequence of activity indicators and flow intensity. The overall stability curve and local activity distribution map are compared with the preset knowledge health benchmark model; Based on the comparison results, rigid nodes and overloaded links in the dynamic knowledge topology network were identified.

8. The intelligent robot knowledge base management method as described in claim 7, characterized in that, The comparison of the overall stability curve and local activity distribution map with the preset knowledge health benchmark model specifically includes: Obtain a preset knowledge health benchmark model, which includes a standard overall stability curve template and a standard local activity distribution map template; The overall stability curve obtained by fitting the evolution analyzer is compared point by point with the standard overall stability curve template, and the sum of squares of the differences between the two curves at each time point is calculated as the overall deviation. The local activity distribution map generated by the evolutionary analyzer is compared with the standard local activity distribution map template by dividing the region, and the cosine similarity of the activity values ​​of the corresponding region is calculated as the local similarity. Based on the overall deviation and local similarity, calculate the comprehensive difference index between the current state of the dynamic knowledge topology network and the knowledge health benchmark model; The overall health status of the dynamic knowledge topology network is determined by comparing the comprehensive difference index with a preset threshold.

9. The intelligent robot knowledge base management method as described in claim 5, characterized in that, After identifying the stagnant nodes and overloaded links in the dynamic knowledge topology network based on the comparison results, the method further includes: A knowledge waking process is initiated for the identified stagnant nodes, which includes injecting new association query requests or association context information into the stagnant nodes. To initiate a traffic diversion process for the identified overloaded links, the traffic diversion process includes building new auxiliary connections near the overloaded links or queuing and scheduling knowledge requests flowing to the overloaded links. The updated status of the stagnant nodes processed by the knowledge awakening process and the overloaded links processed by the traffic diversion process are fed back to the dynamic knowledge topology network to complete the real-time adjustment of the dynamic knowledge topology network.

10. An intelligent robot knowledge base management system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent robot knowledge base management method according to any one of claims 1 to 9.

Images