A power business hall multi-channel business coordination shunting scheduling method and system

Abstract

The application provides a power business hall multi-channel business collaborative shunting scheduling method and system, collects customer business demand, various channel service configurations and real-time running states, respectively constructs a multi-dimensional demand vector with positive and negative components, a same dimension channel feature vector and a channel state vector, decomposes the dot product of the demand vector and the channel feature vector, calculates a matching purity index, corrects the correlation distance with the matching purity by combining the atlas business atomic node semantic distance, represents the conflict influence by means of a conflict penalty coefficient, obtains a numerical matching degree, obtains an initial similarity by fusing the matching degree and the corrected distance, generates a saturation inhibition factor relying on the channel load and the matching purity, modulates the initial similarity to obtain a scheduling score, and allocates the customer to the channel with the highest score, so that intelligent collaborative shunting is realized.

Description

Technical Field

[0001] This application belongs to the field of dispatching, and in particular relates to a method and system for multi-channel business collaborative diversion dispatching in power business halls. Background Technology

[0002] Power companies are continuously promoting the digital transformation of their marketing services. The service model of their business halls has gradually shifted from a single manual counter to a multi-channel collaborative service model encompassing physical windows, intelligent self-service equipment, mobile clients, and customer service hotlines. Customers' needs when handling power-related business are no longer limited to basic information inquiries or electricity bill payments, but have gradually expanded to include complex needs such as new installations and capacity increases, comprehensive energy consultations, and complex fault reporting. Due to differences in hardware and software configurations, service authorization scope, interaction processing efficiency, and business types across different business channels, the overall service resources of business halls are unevenly distributed. In this scenario, accurately identifying the core orientations and constraints in customers' complex business needs and matching them with the service capabilities of each channel with high adaptability has become a crucial foundation for improving the operational efficiency of power business halls and optimizing customer service experience. Existing power business guidance and diversion management typically relies on rule-based scheduling methods such as basic tag matching and polling allocation, which still have shortcomings in deep semantic parsing capabilities and global collaborative measurement. When customer needs involve interdependent conditions or non-standardized service intentions, existing methods struggle to accurately identify and express the positive synergistic and negative conflict relationships between customer needs and channel attributes, easily leading to demand analysis remaining at a superficial label matching level. Simultaneously, the real-time load pressure and operational status of each business channel are constantly changing. Relying solely on static rules for traffic allocation makes it difficult to link supply and demand matching with channel capacity for evaluation. During sudden customer flow peaks or concentrated complex business events, this type of traffic diversion can easily cause scheduling deviations, leading to congestion in some highly matched channels due to concentrated traffic, while other channel resources with the capacity to share the load are not fully utilized, thus hindering the efficiency of multi-channel business collaborative processing in power service halls. Summary of the Invention

[0003] To improve the accuracy of multi-channel resource allocation, the depth of customer demand analysis, and the efficiency of business diversion and coordination in power service halls, this invention provides a method and system for multi-channel business coordination and diversion scheduling in power service halls.

[0004] The first aspect of this invention provides a method for multi-channel business collaborative diversion and scheduling in power service halls, comprising the following steps:

[0005] The system acquires the business requirements of customers to be processed and constructs a multi-dimensional business requirement vector containing positive and negative components based on the power business ontology knowledge graph. It also acquires the service capability configuration of all business channels and constructs a channel feature vector with the same dimension as the multi-dimensional business requirement vector based on the power business ontology knowledge graph. Finally, it acquires the operating status of all business channels and constructs a channel status vector.

[0006] For the multidimensional business demand vector and any of the channel feature vectors, the dot product of the two is decomposed into directional coordination components and directional conflict components to calculate the matching purity index. The positive non-zero components used for semantic association distance calculation in the two vectors are extracted to correspond to the business atomic nodes in the power business ontology knowledge graph. The semantic association distance between the business atomic nodes is calculated. The matching purity index is used as an adjustment factor to correct the semantic association distance to obtain the corrected semantic association distance.

[0007] The directional conflict component is weighted using a preset conflict penalty coefficient. The weighted directional conflict component is subtracted from the directional coordination component to obtain a numerical matching degree. The numerical matching degree is fused with the corrected semantic association distance to obtain an initial similarity score. Based on the load index in the channel state vector and using the matching purity index, the inhibition curve is adjusted to generate a channel saturation inhibition factor. The channel saturation inhibition factor is used to modulate the initial similarity score to obtain a scheduling score. The customers to be processed are diverted to the business channel with the highest scheduling score.

[0008] A second aspect of the present invention provides a multi-channel business collaborative diversion and dispatching system for power service halls, comprising the following modules:

[0009] The module is used to acquire the business requirements of customers to be processed and construct a multi-dimensional business requirement vector containing positive and negative components based on the power business ontology knowledge graph; acquire the service capability configuration of all business channels and construct a channel feature vector with the same dimension as the multi-dimensional business requirement vector based on the power business ontology knowledge graph; and acquire the operating status of all business channels and construct a channel status vector.

[0010] The calculation module is used to decompose the dot product of the multidimensional business demand vector and any of the channel feature vectors into directional coordination components and directional conflict components to calculate the matching purity index, extract the positive non-zero components in the two vectors used for semantic association distance calculation and their corresponding business atomic nodes in the power business ontology knowledge graph, calculate the semantic association distance between the business atomic nodes, and use the matching purity index as an adjustment factor to correct the semantic association distance to obtain the corrected semantic association distance.

[0011] The scheduling module is used to weight the directional conflict component using a preset conflict penalty coefficient, subtract the weighted directional conflict component from the directional coordination component to obtain a numerical matching degree, fuse the numerical matching degree with the corrected semantic association distance to obtain an initial similarity score, generate a channel saturation suppression factor based on the load index in the channel state vector and the matching purity index to adjust the suppression curve, modulate the initial similarity score using the channel saturation suppression factor to obtain a scheduling score, and divert the customers to be processed to the business channel with the highest scheduling score.

[0012] This invention constructs a multi-dimensional vector containing positive and negative components based on a power business ontology knowledge graph to identify power business characteristics by representing customer demand and channel service capabilities. The vector dot product calculation is decomposed into directional coordination and directional conflict components, representing the matching purity between demand and channels. Directional correction is performed by combining the semantic association distance of business atomic nodes, improving the semantic accuracy and logical rigor of similarity calculation. A conflict penalty coefficient is introduced to constrain mismatch factors, and a more scientific channel saturation suppression mechanism is constructed based on the current actual load status of the channels to adjust the initial score. This method integrates numerical matching degree, semantic association, and channel load status to generate scheduling scores, which helps improve the balance of multi-channel resource allocation, alleviate single-channel congestion problems, and improve the overall business processing efficiency of power business halls and the scientific nature of customer collaborative diversion scheduling. Attached Figure Description

[0013] Figure 1 A flowchart of a multi-channel business collaborative diversion and scheduling method for power service halls;

[0014] Figure 2 Flowchart for building and decomposing business-channel vectors;

[0015] Figure 3 This is a schematic diagram of a multi-channel operation indicator and scheduling parameter record table. Detailed Implementation

[0016] The terms "first," "second," and "third," etc., used in this application specification, claims, and the aforementioned drawings are used to distinguish different objects, not to limit a specific order.

[0017] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0018] Example 1

[0019] In Embodiment 1 of the present invention, as Figure 1 As shown, a method for multi-channel business collaborative diversion and scheduling in a power service hall includes the following steps:

[0020] S1. Obtain the business requirements of the customers to be processed and construct a multi-dimensional business requirement vector containing positive and negative components based on the power business ontology knowledge graph. Obtain the service capability configuration of all business channels and construct a channel feature vector with the same dimension as the multi-dimensional business requirement vector based on the power business ontology knowledge graph. Obtain the operating status of all business channels and construct a channel status vector.

[0021] The system acquires natural language business requirements from customers to be processed, performs word segmentation and part-of-speech tagging using a natural language processing toolkit, and extracts business keywords using a pre-trained BERT model for named entity recognition and intent classification. These extracted keywords are input into a power business ontology knowledge graph built on the Neo4j graph database for node matching and querying of entity nodes and relationships. For matched business nodes, positive weights are assigned to affirmative statements in the customer requirements, forming positive components; negative weights are assigned to negative statements or exclusion conditions, forming negative components. An array function is used to initialize and construct a multi-dimensional business requirement vector. Simultaneously, service permission forms from all business channels—manual counters, self-service terminals, and the online app—are read through the business hall management system interface and mapped to business nodes in the Neo4j graph database using the same method, generating a channel feature vector with the same dimensions as the multi-dimensional business requirement vector. The system collects current queue length, average service time per transaction, and total equipment failure interruption time, and obtains the current availability status identifier for each business channel. The collected operational data is normalized and combined with the current availability status identifier to construct a one-dimensional channel status vector.

[0022] In one embodiment, the step of acquiring the business requirements of the customer to be processed and constructing a multi-dimensional business requirement vector containing positive and negative components based on the power business ontology knowledge graph includes:

[0023] Obtain the processing description text submitted by the customer and the urgency level value set by the user;

[0024] The processing description text is segmented and tagged with parts of speech, and special words representing the action object and function are extracted to form a text word set;

[0025] The text word set is mapped to the set of business atomic nodes of the power business ontology knowledge graph. The positive association values ​​of the matching demand business atomic nodes are determined or adjusted according to the urgency value. Negative association values ​​are assigned to the business atomic nodes that are explicitly excluded. The combination forms the multi-dimensional business demand vector corresponding to the node dimension of the knowledge graph.

[0026] The system receives customer-submitted description text, such as "I need to pay my electricity bill, not broadband," along with an urgency level value E set via the interactive interface. The preferred value is an integer within the range [1, 5], where 1 represents normal and 5 represents extremely urgent. This text is segmented and tagged with parts of speech, extracting nouns and verbs. Invalid words are filtered using a pre-defined list of prohibited words for electricity services, resulting in a text word set containing core action objects and functions. Keywords in the text word set are matched to business atomic nodes in a power service ontology knowledge graph with dimension N, for example, N=500. During vector construction, a zero vector of length N is initialized as the initial multidimensional business requirement vector. For each matching business atomic node in the text word set, such as "pay electricity bill," the urgency level value E set by the user on the interactive interface is read, and the positive correlation value of the business atomic node is determined or adjusted based on this value. For example, if E=4, the positive correlation value is set to 0.8. By relying on syntactic analysis to identify words with negative prefixes, such as "not applying for broadband," these explicitly excluded business atomic nodes, such as "broadband application," are assigned fixed negative correlation values, preferably within the range of [-1.0, -0.5], for example, set to -0.8. The absolute value of these negative correlation values ​​can be adjusted according to the user's negation intensity, business exclusion priority, or preset rules. These positive and negative values ​​are then filled into an initial zero vector to generate a multi-dimensional business requirement vector containing positive and negative components, with dimensions corresponding to knowledge graph nodes, thereby realizing the expression of business requirements in the semantic space of the knowledge graph.

[0027] In one embodiment, the steps of obtaining the service capability configuration of all business channels and constructing a channel feature vector with the same dimension as the multidimensional business demand vector based on the power business ontology knowledge graph, and obtaining the operating status of all business channels and constructing a channel status vector, include:

[0028] Extract the business configuration information that is supported and not supported by each business channel, map the information to the corresponding business atomic nodes of the power business ontology knowledge graph, assign positive capability values ​​and negative capability values ​​respectively, and construct the channel feature vector with the same dimension as the multidimensional business demand vector.

[0029] Get the total number of people currently waiting in the queue for each business channel;

[0030] Extract the average service duration of a single transaction and the total downtime due to equipment failure for each business channel in the previous statistical period.

[0031] Obtain the current availability status identifier of each business channel, which is used to indicate whether the corresponding business channel is in a processable state;

[0032] The total number of people currently waiting in the queue, the average service time per transaction, and the total interruption time due to equipment failure are respectively subjected to maximum and minimum normalization processing, and combined with the current available status identifier to form a one-dimensional feature array as the channel status vector.

[0033] Extract a list of supported and unsupported business configurations for various business channels, such as manual counters, self-service terminals (Type A), and online WeChat mini-programs. Map these lists to the corresponding business atomic nodes in a power business ontology knowledge graph with dimension N, for example, N=500. For business nodes supported by a channel, such as self-service terminals supporting "electricity bill payment," assign a positive capability value, preferably set to 1.0; for business nodes explicitly not supported by the channel's hardware or permissions, such as self-service terminals not supporting "new installations for large industrial users," assign a negative capability value, preferably set to -1.0; leave nodes without clear definitions as 0. This generates a channel feature vector with the same dimension as the multi-dimensional business requirement vector. Simultaneously, obtain the total number of people currently waiting in line for each channel through the queuing system interface of the business hall, variable Q, for example, Q=15 people, and extract the average service time per transaction in the past 60 minutes, variable T, for example, T=180 seconds, and the total interruption time due to equipment failure, variable F, for example, F=0 seconds, through the log analysis engine. The three variables are normalized using the historical maximum and minimum values. For example, the historical maximum value is set to... , Second, seconds, after normalization , , The processed Q′, T′, and F′, along with the current available status identifier A, are concatenated in a fixed order into a one-dimensional feature array of length 4, such as [0.3, 0.3, 0.0, 1]. Here, A=1 indicates that the channel is available and A=0 indicates that the channel is not available. This forms a channel status vector for subsequent real-time scheduling, enabling the comparability of the representation of multi-channel load conditions.

[0034] S2, for the multidimensional business demand vector and any of the channel feature vectors, the dot product of the two is decomposed into directional coordination components and directional conflict components to calculate the matching purity index. The positive non-zero components used for semantic association distance calculation in the two vectors are extracted to correspond to the business atomic nodes in the power business ontology knowledge graph. The semantic association distance between the business atomic nodes is calculated. The matching purity index is used as an adjustment factor to correct the semantic association distance to obtain the corrected semantic association distance.

[0035] For the multidimensional business demand vector and the channel feature vector, perform element-wise multiplication of corresponding elements to generate an intermediate result vector, such as... Figure 2As shown, the intermediate result vector is traversed, and the elements obtained by multiplying the positive correlation values ​​in the multidimensional business demand vector with the corresponding positive capability values ​​in the channel feature vector are summed to obtain the directional coordination component. Only the absolute values ​​of the products of the positive correlation values ​​in the multidimensional business demand vector and the corresponding negative capability values ​​in the channel feature vector are summed to obtain the directional conflict component. Negative demand nodes explicitly excluded by the customer and the positive capability values ​​corresponding to the channel's support for that business are not included in the directional conflict component. These explicitly excluded negative demand nodes are used to exclude them from the semantic association distance calculation and initial similarity improvement process, unless the corresponding channel has default bundled processing or forced guidance rules; otherwise, the channel's ability to process that business does not constitute a separate directional conflict component. The difference between the directional coordination component and the directional conflict component is divided by the sum of the directional coordination component, the directional conflict component, and a preset minimum positive number to calculate a matching purity index with a value range of [-1, 1]. The `nonzero` function is called to find the index positions corresponding to the positive non-zero components used for semantic association distance calculation in two vectors. These indexes are then located in the Neo4j graph database, and the Dijkstra path algorithm is used to calculate the shortest weighted path length between the corresponding node pairs. This shortest weighted path length is mapped to the basic semantic association distance. An exponential decay function is used to set the exponent of the base Euler number `e` to a negative matching purity index to generate a weight adjustment coefficient. The basic semantic association distance is multiplied by this weight adjustment coefficient to correct the semantic association distance, and the corrected semantic association distance is output. A higher matching purity index corresponds to a smaller weight adjustment coefficient and a smaller corrected semantic association distance, thus reducing the decay effect of semantic distance on the initial similarity score in high-purity matching scenarios. Nodes with explicitly excluded negative requirements and nodes with negative capabilities not supported by the channel are used for directional conflict or capability constraint calculations and are not considered as semantic association distance nodes to improve the initial similarity score.

[0036] In one embodiment, the step of decomposing the dot product calculation of the two components into directional cooperation components and directional conflict components to calculate the matching purity index includes:

[0037] The sum of the products of the positive correlation values ​​in the multidimensional business demand vector and the positive capability values ​​of the corresponding dimensions in the channel feature vector is used as the directional collaboration component.

[0038] The sum of the absolute values ​​of the products of the positive correlation values ​​in the multidimensional business demand vector and the corresponding negative capability values ​​in the channel feature vector is used as the directional conflict component.

[0039] The difference between the directional cooperative component and the directional conflict component is used as the numerator, and the sum of the directional cooperative component, the directional conflict component, and a preset minimum positive number is used as the denominator.

[0040] The ratio is obtained by dividing the numerator by the denominator, and the ratio is retained to a predetermined number of decimal places as the matching purity index.

[0041] Traverse each dimension i, i∈[1,N], of the multidimensional business demand vector denoted as vector U and the channel feature vector denoted as vector V. When a component is detected... The values ​​are positively correlated and When the value is positive capability, the product of the two values ​​is calculated and summed to obtain the directional cooperative component. For example, the "electricity bill payment" dimension in the demand vector has a value of 0.8, and the corresponding value in the channel vector is 1.0. The product is 0.8. Summing up all dimensions that meet the criteria, let's assume we get... This component represents the degree of positive synergy between customer needs and channel service capabilities. It filters out dimensions that customers positively require but that the channel explicitly does not support, such as the customer needing "new installations for large industrial users." =0.9, while a certain self-service terminal channel has a negative capability value in this dimension. =-1.0, calculate the absolute value of the product of these dimensional components and sum them to obtain the directional conflict component. For example, by accumulating... =0.9. For customers who explicitly refuse the "broadband application" dimension. =-0.8, while this dimension is a supporting item for a certain channel. When the value is 1.0, it is not included in the direction conflict component. The formula for calculating the matching purity index is executed, setting the molecule as... For example, if the result is 1.5, the denominator is set to... For example, if the result is 3.3, a minimum constant will be added to the denominator in the actual calculation. To prevent division by zero anomalies, the numerator is divided by the denominator to obtain a ratio of 0.4545. A truncation algorithm is then used to retain this ratio to a preset number of decimal places, preferably 4 decimal places, i.e., 0.4545. This output serves as a matching purity index for subsequent result confidence assessment. To represent the interference of redundant channel business functions with services explicitly unnecessary to customers, in an alternative embodiment, the sum of the products of the positive correlation values ​​in the multidimensional business demand vector and the corresponding positive capability values ​​in the channel feature vector is calculated as the directional coordination component. The absolute values ​​of the products of the positive correlation values ​​and the corresponding negative capability values ​​in the multidimensional business demand vector, as well as the absolute values ​​of the products of the negative correlation values ​​and the corresponding positive capability values ​​in the same dimension, are calculated separately. The sum of these two absolute values ​​is then used as the directional conflict component.

[0042] In one embodiment, calculating the semantic association distance between the business atomic nodes includes:

[0043] In the power business ontology knowledge graph, the business atomic nodes corresponding to the positive non-zero components used for semantic association distance calculation in the multi-dimensional business demand vector and the channel feature vector are respectively located as the starting node set and the ending node set.

[0044] For each business atomic node in the set of starting nodes, the shortest path algorithm is used to calculate the shortest weighted path length from the business atomic node to each node in the set of ending nodes, and the corresponding minimum path length is selected as the local semantic distance of the business atomic node.

[0045] Based on the positive correlation values ​​of each business atomic node in the multidimensional business demand vector, the weighted average of each local semantic distance is used as the semantic correlation distance between the two groups of business atomic nodes in the knowledge graph.

[0046] Load the pre-built power business ontology knowledge graph, extract the graph nodes corresponding to the positive non-zero components in the multi-dimensional business demand vector, and aggregate the nodes into a starting node set S, such as node S. Representing 'personal electricity bill payment', the business atomic nodes corresponding to the positive non-zero components in the channel feature vector are extracted and aggregated into a set of termination nodes T, such as nodes. The atomic node representing the 'electricity bill payment' business indicates that the corresponding channel supports this business in the channel feature vector. Explicitly excluded negative demand nodes and negative capability nodes not supported by the channel are reflected in the directional conflict component and numerical matching degree. Edges between nodes in the knowledge graph are pre-weighted based on business relevance, typically ranging from [0.1, 1.0], with closer relationships resulting in smaller edge weights. For each node pairing combination in sets S and T, Dijkstra's shortest path algorithm is called to calculate the shortest weighted path length between each node pair. When no reachable path exists between any node pair, the shortest weighted path length of that pair is set to the pre-set maximum semantic distance; when all node pairs are unreachable, the semantic relevance distance between the two sets of business atomic nodes in the knowledge graph is set to the pre-set maximum semantic distance. After traversing and calculating all node pairs between sets S and T a total of |S|×|T| calculations, a distance matching matrix is ​​generated, recording the shortest weighted path length of all candidate matching pairs. For example, node... arrive The shortest weighted path length calculation result is 0.2, while arrive The length is 0.9. The minimum path length from each starting node to the set of ending nodes is extracted using a comparison and filtering algorithm to obtain the local semantic distance for each starting node. This distance is then weighted and averaged according to the positive correlation values ​​of each starting node in the multi-dimensional business demand vector. The weighted average result is assigned as the semantic correlation distance between the two sets of business atomic nodes in the knowledge graph semantic space. The smaller this distance value, the closer the customer's positive demand is to the channel's positive service capabilities in terms of business logic.

[0047] S3, the directional conflict component is weighted using a preset conflict penalty coefficient, and the weighted directional conflict component is subtracted from the directional coordination component to obtain a numerical matching degree. The numerical matching degree is fused with the corrected semantic association distance to obtain an initial similarity score. Based on the load index in the channel state vector and using the matching purity index, the inhibition curve is adjusted to generate a channel saturation inhibition factor. The channel saturation inhibition factor is used to modulate the initial similarity score to obtain a scheduling score. The customers to be processed are diverted to the business channel with the highest scheduling score.

[0048] A conflict penalty coefficient greater than 1 is pre-set. This coefficient is multiplied by the directional conflict component to achieve weighting. Subtraction is then performed to deduct the weighted directional conflict component from the directional coordination component to obtain the numerical matching degree. Numerical matching degrees less than zero are set to zero, while those greater than or equal to zero remain unchanged, resulting in a non-negative numerical matching degree. The distance decay coefficient corresponding to the corrected semantic association distance is calculated, and the non-negative numerical matching degree is multiplied by the distance decay coefficient to obtain the initial similarity score. The current queue length, average service time per transaction, and total equipment failure interruption time are extracted from the channel state vector as load indicators. The dimensionless load indicators are weighted and summed to obtain the comprehensive channel load value. An exponential decay function is constructed as the inhibition curve. The exponential decay function is constructed as follows: Where L represents the channel comprehensive load value, β represents the smoothing adjustment parameter, γ represents the decay rate adjustment parameter obtained from the difference between a preset normal number greater than 1 and the matching purity index, and I represents the channel saturation suppression factor. The difference between the preset normal number greater than 1 and the matching purity index is used as the decay rate adjustment parameter. The channel comprehensive load value is input into the adjusted exponential decay function to calculate the channel saturation suppression factor in the range of 0 to 1. The initial similarity score is multiplied by the channel saturation suppression factor to complete numerical modulation and obtain the scheduling score of each business channel. Before determining the maximum scheduling score, the current available status identifier in the channel status vector is read. When a business channel is in an unavailable state, or its normalized equipment failure interruption index exceeds the preset failure threshold, the scheduling score of that business channel is set to zero, or it is directly removed from the candidate business channel set. The channel index corresponding to the maximum scheduling score among the remaining candidate business channels is found, and the customer's number retrieval information and business demand data are distributed to the terminal processing system corresponding to that channel index to complete collaborative diversion, such as... Figure 3 As shown, each business channel is a candidate business channel that can be processed, indicated by its current availability status.

[0049] In one embodiment, fusing the numerical matching degree with the corrected semantic association distance to obtain an initial similarity score includes:

[0050] The absolute value of the corrected semantic association distance is negative, and the distance decay coefficient is obtained by calculating the exponent with the natural constant as the base.

[0051] Set the matching degree of the numerical values ​​less than zero to zero, and keep the matching degree of the numerical values ​​greater than or equal to zero unchanged to obtain the non-negative numerical matching degree.

[0052] The non-negative numerical matching degree is multiplied by the distance decay coefficient and the output is used as the initial similarity score.

[0053] To obtain the semantic association distance after correction for purity adjustment factor, the variable is denoted as... Next, extract the absolute value of the variable. To ensure the non-negativity of the distance scalar, the absolute value is negativeed to obtain... For example, if the corrected semantic association distance If so, then it is treated as -0.5. Using the base of the natural logarithm, e, which is approximately 2.71828, as the base, ... As an exponent, it performs exponentiation, that is, it calculates... This is used to obtain the distance attenuation coefficient. This exponential transformation smoothly maps the distance value to a continuous interval (0,1], making the attenuation penalty larger the distance, for example, 0.6065 is calculated. After calculating the distance attenuation coefficient, the non-negative numerical matching degree obtained in the previous stage is extracted, and the variable is denoted as . Assuming =1.5, and perform a scalar multiplication operation on this non-negative numerical matching degree and the calculated distance decay coefficient. The product of the two, 0.9098, is output by the system in real time and serves as the initial similarity score for the current customer to be processed by this specific business channel. Through multiplication fusion, while preserving the numerical matching strength in the multidimensional vector space, semantic penalty weights from the knowledge graph topology are injected, improving the rationality of the comprehensive evaluation system.

[0054] In one embodiment, the step of generating a channel saturation suppression factor based on the load index in the channel state vector and adjusting the suppression curve using the matching purity index includes:

[0055] Extract the number of people queuing, the average service time per transaction, and the total downtime due to equipment failure from the channel status vector as load indicators, and extract the current availability status identifier from the channel status vector.

[0056] The channel's overall load value is obtained by weighting and summing the number of people in the queue, the average service time per transaction, and the total downtime of equipment failures after dimensionless processing using preset weighting coefficients.

[0057] The channel comprehensive load value and smoothing adjustment parameter are input into the exponential decay function, and the difference between the preset normal number greater than 1 and the matching purity index is used as the decay rate adjustment parameter of the exponential decay function to calculate the channel saturation suppression factor.

[0058] When the current availability status indicator indicates that the service channel is unavailable, or when the normalized equipment failure interruption index exceeds the preset failure threshold, the scheduling score of the corresponding service channel is set to zero, or the corresponding service channel is removed from the candidate service channel set.

[0059] Three load index variables are extracted from the channel state vector after being processed by max-min dimensionless transformation: the number of people queuing is denoted as... The average service time per transaction is recorded as follows: And the total duration of equipment failure interruption is recorded as Based on actual operational conditions, a normalized weighting coefficient is preset, with the optimal configuration being the weight of the number of people in the queue. Average time consumption weight Fault interruption weight Using the linear weighted summation formula L= ·Q'+ ·T'+ ·F' calculates the channel's overall load value L, for example , , The calculated value is L=0.24. This scalar reflects the resource utilization rate of the channel at this moment. The normalized equipment failure interruption index participates in gradual suppression as part of the overall load when it does not exceed the preset failure threshold. When it exceeds the preset failure threshold, it triggers the corresponding business channel's scheduling score to be reset to zero or the candidate to be removed. The process then proceeds to the stage of generating the channel saturation suppression factor, where a smoothing adjustment parameter is set. The preferred range is [1.5, 3.0]. For example, if we take... and a preset positive constant greater than 1 The preferred range is [1.1, 1.5]. For example, if we take... Set the preset positive constants. Subtracting the purity index P (e.g., P=0.5) from the value of 0.7 yields a difference of 0.7, which is then assigned as the decay rate adjustment parameter of the exponential decay function. The channel saturation inhibition factor is calculated using the exponential decay formula. Substituting the data, we get I≈0.9605. Through this nonlinear adjustment mechanism, the channel with higher matching purity experiences smaller scheduling penalty decay, while high-load congested channels will be subject to corresponding score suppression, achieving synergy between business matching and network resource load balancing.

[0060] Example 2

[0061] Embodiment 2 of the present invention proposes a multi-channel business collaborative diversion and dispatching system for power business halls, including the following modules:

[0062] The module is used to acquire the business requirements of customers to be processed and construct a multi-dimensional business requirement vector containing positive and negative components based on the power business ontology knowledge graph; acquire the service capability configuration of all business channels and construct a channel feature vector with the same dimension as the multi-dimensional business requirement vector based on the power business ontology knowledge graph; and acquire the operating status of all business channels and construct a channel status vector.

[0063] The calculation module is used to decompose the dot product of the multidimensional business demand vector and any of the channel feature vectors into directional coordination components and directional conflict components to calculate the matching purity index, extract the positive non-zero components in the two vectors used for semantic association distance calculation and their corresponding business atomic nodes in the power business ontology knowledge graph, calculate the semantic association distance between the business atomic nodes, and use the matching purity index as an adjustment factor to correct the semantic association distance to obtain the corrected semantic association distance.

[0064] The scheduling module is used to weight the directional conflict component using a preset conflict penalty coefficient, subtract the weighted directional conflict component from the directional coordination component to obtain a numerical matching degree, fuse the numerical matching degree with the corrected semantic association distance to obtain an initial similarity score, generate a channel saturation suppression factor based on the load index in the channel state vector and the matching purity index to adjust the suppression curve, modulate the initial similarity score using the channel saturation suppression factor to obtain a scheduling score, and divert the customers to be processed to the business channel with the highest scheduling score.

[0065] It is understood that the various numerical designations used in the embodiments of this application are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application. The order of the process numbers described above does not imply the order of execution; the execution order of each process should be determined by its function and internal logic.

Claims

1. A method for multi-channel business collaborative diversion and scheduling in a power service hall, characterized in that, Includes the following steps: The system acquires the business requirements of customers to be processed and constructs a multi-dimensional business requirement vector containing positive and negative components based on the power business ontology knowledge graph. It also acquires the service capability configuration of all business channels and constructs a channel feature vector with the same dimension as the multi-dimensional business requirement vector based on the power business ontology knowledge graph. Finally, it acquires the operating status of all business channels and constructs a channel status vector. For the multidimensional business demand vector and any of the channel feature vectors, the dot product of the two is decomposed into directional coordination components and directional conflict components to calculate the matching purity index. The positive non-zero components used for semantic association distance calculation in the two vectors are extracted to correspond to the business atomic nodes in the power business ontology knowledge graph. The semantic association distance between the business atomic nodes is calculated. The matching purity index is used as an adjustment factor to correct the semantic association distance to obtain the corrected semantic association distance. The directional conflict component is weighted using a preset conflict penalty coefficient. The weighted directional conflict component is subtracted from the directional coordination component to obtain a numerical matching degree. The numerical matching degree is fused with the corrected semantic association distance to obtain an initial similarity score. Based on the load index in the channel state vector and using the matching purity index, the inhibition curve is adjusted to generate a channel saturation inhibition factor. The channel saturation inhibition factor is used to modulate the initial similarity score to obtain a scheduling score. The customers to be processed are diverted to the business channel with the highest scheduling score.

2. The method according to claim 1, characterized in that, The process of acquiring the business requirements of customers to be processed and constructing a multi-dimensional business requirement vector containing positive and negative components based on the power business ontology knowledge graph includes: Obtain the processing description text submitted by the customer and the urgency level value set by the user; The processing description text is segmented and tagged with parts of speech, and special words representing the action object and function are extracted to form a text word set; The text word set is mapped to the set of business atomic nodes of the power business ontology knowledge graph. The positive association values ​​of the matching demand business atomic nodes are determined or adjusted according to the urgency value. Negative association values ​​are assigned to the business atomic nodes that are explicitly excluded. The combination forms the multi-dimensional business demand vector corresponding to the node dimension of the knowledge graph.

3. The method according to claim 1, characterized in that, The process of obtaining the service capability configuration of all business channels and constructing a channel feature vector with the same dimension as the multidimensional business demand vector based on the power business ontology knowledge graph, and obtaining the operating status of all business channels and constructing a channel status vector, includes: Extract the business configuration information that is supported and not supported by each business channel, map the information to the corresponding business atomic nodes of the power business ontology knowledge graph, assign positive capability values ​​and negative capability values ​​respectively, and construct the channel feature vector with the same dimension as the multidimensional business demand vector. Get the total number of people currently waiting in the queue for each business channel; Extract the average service duration of a single transaction and the total downtime due to equipment failure for each business channel in the previous statistical period. Obtain the current availability status identifier of each business channel, which is used to indicate whether the corresponding business channel is in a processable state; The total number of people currently waiting in the queue, the average service time per transaction, and the total interruption time due to equipment failure are respectively subjected to maximum and minimum normalization processing, and combined with the current available status identifier to form a one-dimensional feature array as the channel status vector.

4. The method according to claim 2, characterized in that, The step of decomposing the dot product calculation of the two components into directional cooperative components and directional conflict components to calculate the matching purity index includes: The sum of the products of the positive correlation values ​​in the multidimensional business demand vector and the positive capability values ​​of the corresponding dimensions in the channel feature vector is used as the directional collaboration component. The sum of the absolute values ​​of the products of the positive correlation values ​​in the multidimensional business demand vector and the corresponding negative capability values ​​in the channel feature vector is used as the directional conflict component. The difference between the directional cooperative component and the directional conflict component is used as the numerator, and the sum of the directional cooperative component, the directional conflict component, and a preset minimum positive number is used as the denominator. The ratio is obtained by dividing the numerator by the denominator, and the ratio is retained to a predetermined number of decimal places as the matching purity index.

5. The method according to claim 1, characterized in that, The calculation of the semantic association distance between the business atomic nodes includes: In the power business ontology knowledge graph, the business atomic nodes corresponding to the positive non-zero components used for semantic association distance calculation in the multi-dimensional business demand vector and the channel feature vector are respectively located as the starting node set and the ending node set. For each business atomic node in the set of starting nodes, the shortest path algorithm is used to calculate the shortest weighted path length from the business atomic node to each node in the set of ending nodes, and the corresponding minimum path length is selected as the local semantic distance of the business atomic node. Based on the positive correlation values ​​of each business atomic node in the multidimensional business demand vector, the weighted average of each local semantic distance is used as the semantic correlation distance between the two groups of business atomic nodes in the knowledge graph.

6. The method according to claim 1, characterized in that, The process of fusing the numerical matching degree with the corrected semantic association distance to obtain the initial similarity score includes: The absolute value of the corrected semantic association distance is negative, and the distance decay coefficient is obtained by calculating the exponent with the natural constant as the base. Set the matching degree of the numerical values ​​less than zero to zero, and keep the matching degree of the numerical values ​​greater than or equal to zero unchanged to obtain the non-negative numerical matching degree. The non-negative numerical matching degree is multiplied by the distance decay coefficient and the output is used as the initial similarity score.

7. The method according to claim 1, characterized in that, The process of generating a channel saturation suppression factor based on the load index in the channel state vector and adjusting the suppression curve using the matching purity index includes: Extract the number of people queuing, the average service time per transaction, and the total downtime due to equipment failure from the channel status vector as load indicators, and extract the current availability status identifier from the channel status vector. The channel's overall load value is obtained by weighting and summing the number of people in the queue, the average service time per transaction, and the total downtime of equipment failures after dimensionless processing using preset weighting coefficients. The channel comprehensive load value and smoothing adjustment parameter are input into the exponential decay function, and the difference between the preset normal number greater than 1 and the matching purity index is used as the decay rate adjustment parameter of the exponential decay function to calculate the channel saturation suppression factor. When the current availability status indicator indicates that the service channel is unavailable, or when the normalized equipment failure interruption index exceeds the preset failure threshold, the scheduling score of the corresponding service channel is set to zero, or the corresponding service channel is removed from the candidate service channel set.

8. A multi-channel business collaborative dispatching system for power service halls, characterized in that, Includes the following modules: The module is used to acquire the business requirements of customers to be processed and construct a multi-dimensional business requirement vector containing positive and negative components based on the power business ontology knowledge graph; acquire the service capability configuration of all business channels and construct a channel feature vector with the same dimension as the multi-dimensional business requirement vector based on the power business ontology knowledge graph; and acquire the operating status of all business channels and construct a channel status vector. The calculation module is used to decompose the dot product of the multidimensional business demand vector and any of the channel feature vectors into directional coordination components and directional conflict components to calculate the matching purity index, extract the positive non-zero components in the two vectors used for semantic association distance calculation and their corresponding business atomic nodes in the power business ontology knowledge graph, calculate the semantic association distance between the business atomic nodes, and use the matching purity index as an adjustment factor to correct the semantic association distance to obtain the corrected semantic association distance. The scheduling module is used to weight the directional conflict component using a preset conflict penalty coefficient, subtract the weighted directional conflict component from the directional coordination component to obtain a numerical matching degree, fuse the numerical matching degree with the corrected semantic association distance to obtain an initial similarity score, generate a channel saturation suppression factor based on the load index in the channel state vector and the matching purity index to adjust the suppression curve, modulate the initial similarity score using the channel saturation suppression factor to obtain a scheduling score, and divert the customers to be processed to the business channel with the highest scheduling score.

9. The system according to claim 8, characterized in that, The process of acquiring the business requirements of customers to be processed and constructing a multi-dimensional business requirement vector containing positive and negative components based on the power business ontology knowledge graph includes: Obtain the processing description text submitted by the customer and the urgency level value set by the user; The processing description text is segmented and tagged with parts of speech, and special words representing the action object and function are extracted to form a text word set; The text word set is mapped to the set of business atomic nodes of the power business ontology knowledge graph. The positive association values ​​of the matching demand business atomic nodes are determined or adjusted according to the urgency value. Negative association values ​​are assigned to the business atomic nodes that are explicitly excluded. The combination forms the multi-dimensional business demand vector corresponding to the node dimension of the knowledge graph.

10. The system according to claim 8, characterized in that, The process of obtaining the service capability configuration of all business channels and constructing a channel feature vector with the same dimension as the multidimensional business demand vector based on the power business ontology knowledge graph, and obtaining the operating status of all business channels and constructing a channel status vector, includes: Extract the business configuration information that is supported and not supported by each business channel, map the information to the corresponding business atomic nodes of the power business ontology knowledge graph, assign positive capability values ​​and negative capability values ​​respectively, and construct the channel feature vector with the same dimension as the multidimensional business demand vector. Get the total number of people currently waiting in the queue for each business channel; Extract the average service duration of a single transaction and the total downtime due to equipment failure for each business channel in the previous statistical period. Obtain the current availability status identifier of each business channel, which is used to indicate whether the corresponding business channel is in a processable state; The total number of people currently waiting in the queue, the average service time per transaction, and the total interruption time due to equipment failure are respectively subjected to maximum and minimum normalization processing, and combined with the current available status identifier to form a one-dimensional feature array as the channel status vector.

Images