An industrial internet transaction path priority construction method and system of an analog neural network

Abstract

The application discloses an industry internet transaction path priority construction method and system of a simulated neural network, and applies a two-stage memory consolidation mechanism of a hippocampus of a brain to optimization of an industry internet transaction path, carries out multidimensional importance evaluation on all business events of a business subject in an active period of business activities, in a dormant period of the business activities, carries out high-frequency compressed playback on business events of the business subject in the active period of the business activities which meet the requirement of importance level, through simulation of a 'wakeful labeling + sleep playback' mechanism of the brain, can automatically identify important business events, and through compressed playback, establishes efficient path connection, significantly improves the intelligent level of a transaction path network, and then recommends the best transaction path for a user, and improves the accuracy of a model.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for constructing an optimal industrial internet transaction path using simulated neural networks. Background Technology

[0002] With the rapid development of the digital economy, the industrial internet has become a crucial infrastructure for driving the digital transformation of the real economy. Unlike the consumer internet, which primarily serves individual users, the industrial internet focuses on connecting enterprises, supply chains, and industrial clusters, achieving intelligent collaboration within the industrial ecosystem through the deep integration of data elements. However, existing industrial internet platforms still have significant shortcomings in path optimization and decision support. Traditional path optimization methods are mainly based on static graph theory algorithms or simple heuristic rules, resulting in inaccurate recommendations, large data processing volumes, low processing efficiency, and high computational costs. Summary of the Invention

[0003] In view of this, the present invention provides a method and system for constructing an optimal transaction path for the industrial internet by simulating a neural network. By simulating the dynamic reconstruction mechanism of the brain's neural network, it can recommend the best transaction path for users.

[0004] A method for constructing an industrial internet transaction path optimization system using simulated neural networks, specifically including the following steps: S1, construct the transaction path network of the operating entity based on the operating entity's historical operating data and the historical supply chain data of all its partners; S2 assesses the importance of all business events that occur during the active period of business operations of the business entity from multiple dimensions. S3, During the dormant period of business activities, the compression replay algorithm is used to compress and replay business events that meet the importance level requirements during the active period of business activities at a high frequency. S4. Based on the replay results of each compressed replay of a business event and the multi-dimensional importance assessment results of that business event, calculate the path optimization probability density of the transaction path involved in each compressed replay of a business event. S5, optimize the transaction path network based on the path optimization probability density of the transaction paths involved in each compressed replay of the business event; S6. Repeat steps S2-S5 to achieve closed-loop continuous optimization of the transaction path network.

[0005] Preferably, in step S1, the specific steps for constructing the basic network of the business entity's transaction path based on the business entity's historical operating data and the historical supply chain data of all its partners are as follows: Collect historical operating data of business entities within a set time window, and obtain supply chain data of all their partners within the same time window through API interfaces; The business entity and its partners are treated as network nodes, and historical transaction relationships are treated as directed edges, forming a transaction path network for the business entity. The weight of each directed edge is initialized as a weighted value of historical transaction frequency and transaction amount.

[0006] Preferably, in step S2, the specific steps for conducting a multi-dimensional importance assessment of all business events occurring during the active period of business operations are as follows: Real-time collection of operational data for all business events occurring during the active operating period of the business entity, as well as current supply chain data of relevant partners; The multidimensional importance score of each business event is calculated based on the multidimensional importance assessment function, and the importance level is assigned according to the score. Cache the operational data and importance assessment data for each operational event.

[0007] Preferably, the calculation formula for the multi-dimensional importance evaluation function is as follows:

[0008] Where w_k is the importance weight of the kth business event in the mth dimension, f_k is the labeling function of the kth business event in the mth dimension, E_k(t) is the occurrence intensity of the kth business event in the mth dimension at time t, and m is the number of dimensions.

[0009] Preferably, the specific steps in step S3 for using the compression replay algorithm to perform high-frequency compression replay of business events that meet the importance level requirements during the active period of business operations are as follows: Based on the importance level of all business events that occur during the active period of business operations, the business data of business events that meet the importance level requirements are extracted, and the extracted business events are used as replay events. Using a compression replay algorithm, the replay events are replayed at a set frequency with a time compression ratio of 10-100 times. During the replay process, the transaction paths involved in the replay events are recorded to obtain the event-path correlation.

[0010] Preferably, the calculation formula for the compressed replay algorithm is:

[0011] Where K(t-τ) is the replay function, I(x, τ) is the importance score of the replay event, and H(τ) is the dormancy exponential function, where H(τ)=1 represents the dormancy period and H(τ)=0 represents the active period. .

[0012] Preferably, in step S4, the specific steps for calculating the path optimization probability density of the transaction paths involved in the compressed replay of the business event are as follows: Calculate the network connection strength of a transaction path based on the event-path association of a certain replay event; The initiator of the transaction in the replay event is taken as the fore-synapse and the receiver as the after-synapse. The weighted comprehensive scoring method is used to calculate the intensity of the fore-synapse activity and the intensity of the after-synapse activity under the combined influence of the three dimensions of business activity, network connection, and decision response. The degree of change in synapse intensity is calculated based on the intensity of the fore-synapse activity and the intensity of the after-synapse activity. The path optimization probability density of the transaction path involved in the replay event is calculated based on the multi-dimensional importance score, compressed replay value, network connection strength and synaptic strength change degree of the transaction path involved in the replay event.

[0013] Preferably, the formula for calculating the path optimization probability density of the transaction paths involved in the replay event is:

[0014] Wherein, P(x,t) is the path optimization probability density of the transaction path involved in the replay event, R(x,t) is the compressed replay value of the replay event, I(x,t) is the multi-dimensional importance score of the replay event, C(x,t) is the congestion degree of transaction traffic or data traffic at time t and transaction path x, α is the path diffusion coefficient of the transaction path involved in the replay event, β is the replay enhancement coefficient of the replay event, γ is the network connection strength coefficient of the transaction path involved in the replay event, and δ is the path attenuation coefficient of the transaction path involved in the replay event. The formula for calculating the network connection strength of the transaction path involved in the replay event is as follows:

[0015] Where W0 is the basic connection strength of the transaction path involved in the replay event, d(i,j) is the distance between the initiating node and the receiving node in the replay event, λ is the connection attenuation length of the transaction path involved in the replay event, η is the transaction frequency enhancement factor between the initiating node and the receiving node in the replay event, and N(i,j) is the number of historical transactions between the initiating node and the receiving node in the replay event. The formula for calculating the degree of change in synaptic strength is:

[0016] Where S(i,j,t) is the degree of change in synaptic strength, A is the learning rate, τ is the synaptic decay time constant, pre(t) is the presynaptic activity intensity, and post(t-Δt) is the postsynaptic activity intensity.

[0017] The formula for calculating the presynaptic activity intensity pre(t) is as follows:

[0018] Wherein, l1(t) is the activity intensity of the presynapse at time t affected by the business activity dimension, l2(t) is the activity intensity of the presynapse at time t affected by the network connection dimension, l3(t) is the activity intensity of the presynapse at time t affected by the decision response dimension, W1 is the weight coefficient of the business activity dimension, W2 is the weight coefficient of the network connection dimension, and W3 is the weight coefficient of the decision response dimension. The formula for calculating the postsynaptic activity intensity post(t-Δt) is as follows:

[0019] Where j1(t-Δt) is the activity intensity of the postsynapse at time t-Δt affected by the business activity dimension, j2(t-Δt) is the activity intensity of the postsynapse at time t-Δt affected by the network connection dimension, j3(t-Δt) is the activity intensity of the postsynapse at time t-Δt affected by the decision response dimension, V1 is the weight coefficient of the business activity dimension, V2 is the weight coefficient of the network connection dimension, V3 is the weight coefficient of the decision response dimension, and Δt is the time delay parameter.

[0020] Preferably, in step S5, the specific steps for optimizing the transaction path network based on the path optimization probability density of the transaction paths involved in each compressed and replayed business event are as follows: Based on the path optimization probability density of the transaction paths involved in each compressed replay of an operational event, retain the transaction paths involved in operational events with a path optimization probability density greater than a set threshold, weaken the network connection strength of the transaction paths involved in operational events with a path optimization probability density less than a set threshold, or delete the transaction paths involved in operational events with a path optimization probability density less than a set threshold.

[0021] Preferably, the method further includes step S7, where the transaction path network recommends the best transaction path to the user.

[0022] Preferably, the method further includes step S8, which involves statistically analyzing the adoption rate, transaction cost, and resource utilization rate of the best transaction path recommended by the transaction path network within a set period, and dynamically adjusting the replay enhancement coefficient β, the network connection strength coefficient γ of the transaction path, and the importance weight w_k of different business events in the importance assessment under different dimensions.

[0023] A system for constructing an optimal transaction path for the industrial internet, simulating a neural network, includes an important event annotation module, a neural dynamics replay module, a path optimization module, an adaptive path construction module, and a visualization representation module. The important event labeling module is used to conduct multi-dimensional importance assessments of all business events that occur during the active period of business operations of an entity, and to classify them into importance levels and label them according to the assessment results; The neurodynamic replay module is used to compress and replay business events that meet the importance level requirements during the active business period using a compression replay algorithm during the dormant business period. The path optimization module is used to calculate the path optimization probability density of the transaction paths involved in each compressed replay of a business event based on the replay results of each compressed replay of the business event and the multi-dimensional importance assessment results of the business event. The adaptive path construction module is used to construct the transaction path network of the business entity based on the business entity's historical operating data and the historical supply chain data of all its partners, and to optimize the transaction path network based on the path optimization probability density of the transaction paths involved in each compressed replay of the business event. The visualization module is used to show the construction process of the transaction path network, the event distribution and importance level of each business event marked with an importance level, and the compressed replay process of the neurodynamic replay module.

[0024] Preferably, it also includes a parameter feedback optimization module, which is used to statistically analyze the adoption rate, transaction cost, and resource utilization rate of the best transaction path recommended by the transaction path network by users within a set period, and to dynamically adjust the replay enhancement coefficient β, the network connection strength coefficient γ of the transaction path, and the importance weight w_k of different business events in importance assessment under different dimensions.

[0025] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0026] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0027] The beneficial effects of this invention are: 1. This invention applies the two-stage memory consolidation mechanism of the hippocampus to the optimization of transaction paths in the industrial internet. By conducting multi-dimensional importance assessments of all business events occurring during the active period of business operations, and during the dormant period of business operations, a compression replay algorithm is used to perform high-frequency compression replay of business events that meet the importance level requirements during the active period of business operations. By simulating the brain's "awake labeling + sleep replay" mechanism, important business events can be automatically identified, and efficient path connections can be established through compression replay, significantly improving the intelligence level of the transaction path network, thereby recommending the best transaction path to users and improving the accuracy of the model.

[0028] 2. In simulating the brain's "awake labeling + sleep replay" mechanism, this invention comprehensively considers multiple factors such as path diffusion, replay enhancement, network connectivity, and path decay, which can accurately describe the dynamic behavior of business entities in the industrial internet, thereby ensuring the accuracy of the model's recommendation results.

[0029] 3. This invention can collect real-time data at the millisecond level, annotate the importance of events at the minute level, replay important events at the hour level, optimize the transaction path network at the daily level, and reconstruct the transaction path network at the weekly level. Through this spectral reinforcement learning mechanism, it integrates multi-timescale learning, bio-inspired memory mechanisms, and adaptive exploration strategies to achieve a more flexible and efficient learning process, enabling it to adapt to complex and ever-changing application scenarios, thereby improving the accuracy of its recommendation results and shortening the response time of its output results.

[0030] 4. Experimental results show that the present invention can achieve significant improvements in several key indicators, such as a 68% reduction in response time, a 47% reduction in transaction costs, a 31% increase in path accuracy, a 37% increase in resource utilization, and an 18% increase in customer satisfaction.

[0031] 5. This invention can be applied to industrial internet application scenarios of different scales and types, and can effectively support everything from departmental applications within an enterprise to regional networks on a global scale, exhibiting good scalability and adaptability.

[0032] 6. The visualization module of this invention can provide an intuitive graphical interface, allowing users to monitor the system's operating status in real time and observe the path optimization process of the transaction path network in real time, so as to evaluate system performance and provide strong support for users to make transaction decisions. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is the overall architecture diagram of the system of the present invention.

[0035] Figure 2 This is a schematic diagram of the important event annotation module.

[0036] Figure 3 This is a schematic diagram of the neurodynamic replay module.

[0037] Figure 4 This is a schematic diagram of the path optimization module.

[0038] Figure 5 This is a schematic diagram of the adaptive path building module.

[0039] Figure 6 This is a user interface diagram of the visualization module.

[0040] Figure 7 This is a schematic diagram of the multi-timescale learning mechanism of the spectral reinforcement learning engine.

[0041] Figure 8 This is a comparison chart of system performance evaluation. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention is described below with reference to specific embodiments shown in the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0043] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0044] To better understand the technical solution of the present invention, the present invention will be described in detail below with reference to the accompanying drawings.

[0045] Example 1: This invention provides a method for constructing an optimal industrial internet transaction path using a simulated neural network, specifically including the following steps: S1 constructs the transaction path network of the operating entity based on the operating entity's historical operating data and the historical supply chain data of all its partners.

[0046] Specifically, the steps for constructing the transaction path network of business entities are as follows: Collect historical operating data of the business entity within a set time window (such as the business entity's transaction records, partner information, order fulfillment data, etc. in the past 6-12 months), and obtain the supply chain data of all its partners in the same time window through API interface (such as the inventory level, delivery capacity, credit rating, etc. of all partners of the business entity in the past 6-12 months). The business entity and its partners are treated as network nodes, and historical transaction relationships are treated as directed edges, forming a transaction path network for the business entity. The weight of each directed edge is initialized as a weighted value of historical transaction frequency and transaction amount.

[0047] Assuming that the initial transaction path network of the formed business entities has N network nodes (usually 100-10000 business entities) and E edges (with a typical length of 500-50000 transaction relationships), then the network density ρ=E / (N*(N-1)).

[0048] S2 assesses the importance of all business events that occur during the active period of a business entity's operations from multiple dimensions.

[0049] The term "active business period" refers to the normal business hours of the business entity, such as 9:00-18:00 on weekdays, during which transaction activities are frequent.

[0050] Specifically, the steps for multi-dimensional importance assessment are as follows: First, real-time data is collected on all business events that occur during the active business period of the business entity, as well as the current supply chain data of relevant partners. The business data of business events that occur during the active business period is collected at the millisecond level, and the collected transaction volume data is statistically analyzed at the minute level. The supply chain data of relevant partners and market data are collected at the second level. Then, the multi-dimensional importance score for each business event is calculated according to the multi-dimensional importance assessment function. The formula for the multi-dimensional importance assessment function is as follows:

[0051] Where w_k is the importance weight of the k-th business event under the m-th dimension, f_k is the labeling function of the k-th business event under the m-th dimension, E_k(t) is the occurrence intensity of the k-th business event under the m-th dimension at time t, and M is the number of dimensions; the so-called "multi-dimensional" includes four dimensions: transaction amount, partner level, market influence, and long-term value. The importance weight of the transaction amount dimension is set to 0.3-0.4, the importance weight of the partner level dimension is set to 0.2-0.3, the importance weight of the market influence dimension is set to 0.2-0.3, and the importance weight of the long-term value dimension is set to 0.1-0.2. Then, each business event is classified into importance levels based on its multi-dimensional importance score, and the importance level is labeled accordingly. In this embodiment, the importance level is classified into level A (high importance, I≥80), level B (medium importance, 60≤I<80), level C (low importance, 40≤I<60), and level D (ignorable, I<40) based on the I(x,t) score. The system caches operational data and importance assessment data for each operational event. The importance assessment data includes multi-dimensional importance scores and importance levels.

[0052] S3, during the dormant period of business operations, uses a compression replay algorithm to perform high-frequency compression replay of business events that meet the importance level requirements that occurred during the active period of business operations.

[0053] The so-called business dormancy period refers to the low period of business activity for the business entity, such as nighttime, weekends, and holidays, during which transaction activity is scarce.

[0054] Specifically, the steps for compressed replay during the dormancy period are as follows: First, based on the importance level of all business events that occurred during the active period of business operations, the system extracts the business data of business events that meet the importance level requirements and uses the extracted business events as replay events. For example, if business events with an importance level of B or above are set to be replayable, the system will automatically extract the business data of business events marked as A and B from all business events that occurred during the active period of business operations. Then, the replay event is replayed using a compression replay algorithm at a time compression ratio of 10-100 times and a set frequency (such as 100-200Hz). During the replay, the transaction path (i.e. which node to which node) involved in the replay event is recorded to obtain the event-path association.

[0055] The calculation formula for the compressed replay algorithm is as follows:

[0056] Where K(t-τ) is the replay function, simulating the sharp wave ripple mechanism of the hippocampus in the brain; I(x, τ) is the importance score of the replay event; and H(τ) is the dormancy exponential function, where H(τ)=1 represents the dormancy period and H(τ)=0 represents the active period. .

[0057] S4. Based on the replay results of each compressed replay of a business event and the multi-dimensional importance assessment results of that business event, calculate the path optimization probability density of the transaction path involved in each compressed replay of a business event.

[0058] Specifically, the steps for calculating the path optimization probability density of the transaction paths involved in the compressed replay of business events are as follows: First, calculate the network connection strength of the transaction path based on the event-path association of a certain replay event; The formula for calculating the network connection strength of the transaction path involved in the replay event is as follows:

[0059] Where W0 is the basic connection strength of the transaction path involved in the replay event, d(i,j) is the distance between the operator and the partner in the replay event, λ is the connection attenuation length of the transaction path involved in the replay event, η is the transaction frequency enhancement factor between the operator and the partner in the replay event, and N(i,j) is the number of historical transactions between the operator and the partner in the replay event.

[0060] Then, the initiator of the transaction in the replay event is designated as the pre-synapse, and the recipient as the post-synapse (for example, if the transaction in the replay event is initiated by the business entity, the business entity is designated as the pre-synapse, and the partner as the post-synapse; if the transaction in the replay event is initiated by the business entity's partner, the partner (such as a supplier) is designated as the pre-synapse, and the business entity as the post-synapse). A weighted comprehensive scoring method is used to calculate the pre-synapse activity intensity and post-synapse activity intensity under the combined influence of the three dimensions: business activity dimension, network connectivity dimension, and decision response dimension. The formula for calculating the presynaptic activity intensity pre(t) is:

[0061] Wherein, l1(t) represents the activity intensity of the presynapse at time t influenced by the business activity dimension, l2(t) represents the activity intensity of the presynapse at time t influenced by the network connection dimension, l3(t) represents the activity intensity of the presynapse at time t influenced by the decision response dimension, W1 represents the weighting coefficient of the business activity dimension, W2 represents the weighting coefficient of the network connection dimension, and W3 represents the weighting coefficient of the decision response dimension. In this embodiment, the weighting coefficient W1 of the business activity dimension is set to 0.5, the weighting coefficient W2 of the network connection dimension is set to 0.3, and the weighting coefficient W3 of the decision response dimension is set to 0.2. Of course, the weighting coefficients of the three dimensions can also be dynamically adjusted.

[0062] The presynaptic activity intensity pre(t) represents the activity level of the initiating node at time t when the replay event occurs. When calculating the presynaptic activity intensity pre(t), it is necessary to first calculate the activity intensities l1(t), l2(t), and l3(t) of the initiating node in the three dimensions of business activity, network connection, and decision response, respectively.

[0063] The activity intensity l1(t) of the initiating node in the business activity dimension is comprehensively affected by three factors: transaction frequency, transaction amount, and order fulfillment rate, and is used to measure the node's business operation activity. Transaction frequency refers to the number of transactions T_freq(t) initiated per unit time, transaction amount refers to the total transaction amount T_amount(t) per unit time, and order fulfillment rate refers to the proportion of orders completed on time (Frate(t)). The formula for calculating the activity intensity l1(t) in the business activity dimension is as follows:

[0064] In the formula, normalize() is the normalization function, which maps indicators of different dimensions to the interval [0,1], wBA1 is the weight coefficient of transaction frequency, wBA2 is the weight coefficient of transaction amount, and w_BA3 is the weight coefficient of order fulfillment rate.

[0065] The activity intensity l2(t) of the initiating node in the network connectivity dimension is comprehensively affected by three factors: out-degree, total connectivity strength, and connectivity activity. It is used to measure the node's connectivity activity in the network. Out-degree refers to the number of edges that the initiating node connects to, Out_degree(t). The total connectivity strength is the sum of the weights of all outgoing edges. Connection activity refers to the proportion of recently activated edges in a network connection dimension, specifically Active_ratio(t). The formula for calculating the activity intensity l2(t) of the network connection dimension is: In the formula, The weighting coefficient for the out-degree. w_NC3 is the weighting factor for the sum of connection strengths, and w_NC3 is the weighting factor for connection activity.

[0066] The activity intensity l3(t) of the initiating node in the decision response dimension is comprehensively affected by three factors: response time, decision accuracy, and strategy adjustment frequency. It is used to measure the speed and quality of the node's decision-making response. Response time refers to the average time T_response(t) from receiving a request to making a response; decision accuracy refers to the accuracy of matching the decision with market demand (Acc_decision(t)); and strategy adjustment frequency refers to the frequency of adjusting the strategy according to market changes (F_adjust(t)). The formula for calculating the activity intensity l3(t) in the decision response dimension is: In the formula, The weighting factor for response time. w_DR3 is the weighting coefficient for decision accuracy and w_DR4 is the weighting coefficient for policy adjustment frequency.

[0067] Based on the activity intensities l1(t), l2(t), and l3(t) of the initiating node in the three dimensions of business activity, network connection, and decision response, and their respective weight coefficients, the presynaptic activity intensity pre(t) can be calculated.

[0068] The formula for calculating the postsynaptic activity intensity post(t-Δt) is:

[0069] Where j1(t-Δt) is the activity intensity of the postsynapse at time t-Δt affected by the business activity dimension, j2(t-Δt) is the activity intensity of the postsynapse at time t-Δt affected by the network connection dimension, j3(t-Δt) is the activity intensity of the postsynapse at time t-Δt affected by the decision response dimension, V1 is the weight coefficient of the business activity dimension, V2 is the weight coefficient of the network connection dimension, V3 is the weight coefficient of the decision response dimension, and Δt is the time delay parameter.

[0070] The postsynaptic activity intensity post(t-Δt) represents the activity level of the receiver node at time (t-Δt); Δt is a time delay parameter, i.e., the duration between the time t when the initiating node initiates the transaction and the time when the receiver node responds to the transaction. The calculation method for post(t-Δt) is exactly the same as that for presynaptic activity, except that the calculation object for presynaptic activity is changed to the receiver node, and the time point is changed from t to (t-Δt). Therefore, the formula for calculating the receiver node's activity intensity j1(t-Δt) in the business activity dimension is: In the formula, The weighting coefficient for transaction frequency. w_BA3 is the weighting factor for transaction amount, and w_BA3 is the weighting factor for order fulfillment rate.

[0071] The formula for calculating the activity intensity j2(t-Δt) of the receiver node in the network connection dimension is: In the formula, The weighting coefficient for the out-degree. w_NC3 is the weighting factor for the sum of connection strengths, and w_NC3 is the weighting factor for connection activity.

[0072] The formula for calculating the activity intensity j3(t-Δt) of the receiver node in the decision response dimension is: In the formula, The weighting factor for response time. w_DR3 is the weighting coefficient for decision accuracy and w_DR4 is the weighting coefficient for policy adjustment frequency.

[0073] The postsynaptic activity intensity post(t-Δt) can be calculated based on the activity intensities j1(t-Δt), j2(t-Δt), and j3(t-Δt) of the receiver node in the three dimensions of business activity, network connection, and decision response, as well as their respective weight coefficients.

[0074] If the presynaptic activity pre(t) occurs shortly after the postsynaptic activity intensity post(t-Δt) (i.e., Δt is positive and small), it indicates that the activity of the initiating node led to the activity of the receiving node. In this case, the transaction path between the two should be strengthened, that is, the edge connecting the initiating node and the receiving node should be strengthened. Conversely, if the order of the presynaptic activity pre(t) and the postsynaptic activity intensity post(t-Δt) is reversed or the time interval between their occurrence is too long, the edge connecting the initiating node and the receiving node should not be strengthened or should be weakened.

[0075] After calculating the presynaptic activity pre(t) and postsynaptic activity intensity post(t-Δt) using the above method, the degree of change in synaptic intensity is calculated based on the presynaptic activity intensity and the postsynaptic activity intensity.

[0076] The formula for calculating the degree of change in synaptic strength is:

[0077] Where S(i,j,t) is the degree of change in synaptic strength, A is the learning rate, τ is the synaptic decay time constant, pre(t) is the presynaptic activity intensity, and post(t-Δt) is the postsynaptic activity intensity.

[0078] Finally, the path optimization probability density of the transaction path involved in the replay event is calculated based on the multi-dimensional importance score I(x,t), the compressed replay value R(x,t), the network connection strength W(i,j) of the transaction path involved in the replay event, and the degree of synaptic strength change S(i,j,t).

[0079] The formula for calculating the path optimization probability density of the transaction paths involved in the replay event is as follows:

[0080] Wherein, P(x,t) is the path optimization probability density of the transaction path involved in the replay event, that is, the probability density of selecting this path at time t and location x (business domain); R(x,t) is the compressed replay value of the replay event; I(x,t) is the multi-dimensional importance score of the replay event; C(x,t) is the congestion degree of transaction traffic or data traffic at time t and transaction path x; α is the path diffusion coefficient of the transaction path involved in the replay event; β is the replay enhancement coefficient of the replay event; γ is the network connection strength coefficient of the transaction path involved in the replay event; and δ is the path attenuation coefficient of the transaction path involved in the replay event.

[0081] C(x,t) represents a factor that negatively impacts or inhibits the probability of a transaction path being optimal; it refers to the economic or resource cost incurred in maintaining or using a particular transaction path. The higher the cost, the lower the probability of the path being optimal. Specifically, C(x,t) represents the congestion level of transaction or data traffic on path x at a specific time t. The more congested the path, the lower its efficiency, and therefore the lower its optimality. The number of nodes included in the path, the complexity of the transaction processes, etc., are also inhibiting factors. Overly complex paths may bring higher risks and management costs, thus being inhibited by the system.

[0082] The C(x,t) parameter, together with the path decay coefficient δ, constitutes a "penalty term" for path selection, ensuring that the system considers not only positive factors (such as replay value and network connectivity strength) but also fully takes into account the inherent negative attributes when evaluating a path.

[0083] In the path optimization probability density formula and its related parameters, x and t represent the following meanings: t (time dimension) t represents time, i.e., the moment when the business event or replay event occurs. In the neurodynamic model of this patent, time is a core variable because the entire system simulates the brain's "wake-up labeling + sleep replay" mechanism, involving the alternation of active and dormant periods of business activity.

[0084] x (Spatial / Business Domain Dimension) 'x' represents a location or business area, used to identify a specific transaction path or business scenario. In the context of the industrial internet, 'x' does not refer to a physical spatial location, but rather to:

[0085] The functions P(x, t), R(x, t), I(x, t), and C(x, t) are all functions concerning "which path / business domain (x)" and "at what time (t)", reflecting the spatiotemporal dynamic characteristics of this patented model. This design enables the system to dynamically calculate and adjust the optimal probability of each transaction path according to different times and different business scenarios.

[0086] S5. Optimize the transaction path network based on the path optimization probability density of the transaction paths involved in each compressed replay of the business event.

[0087] Specifically, based on the path optimization probability density of the transaction paths involved in each compressed replay of an operational event, the transaction paths involved in operational events with a path optimization probability density greater than a set threshold are retained, the network connection strength of the transaction paths involved in operational events with a path optimization probability density less than a set threshold is weakened, or the transaction paths involved in operational events with a path optimization probability density less than a set threshold are deleted.

[0088] S6. Repeat steps S2-S5 to achieve closed-loop continuous optimization of the transaction path network.

[0089] The event granularity of this application includes millisecond-level real-time data acquisition, minute-level event importance labeling, hourly event replay cycle, daily path optimization, and weekly network reconstruction.

[0090] Weekly network reconstruction refers to a comprehensive inspection of the entire transaction path network every week or month, identifying and deleting long-term inactive transaction paths, and discovering potential valuable connections and updating the network based on collaborative filtering algorithms or graph neural networks.

[0091] This invention also provides an industrial internet transaction path optimization and construction system that simulates a neural network, including an important event annotation module, a neural dynamics replay module, a path optimization module, an adaptive path construction module, and a visualization representation module. The important event labeling module is used to conduct multi-dimensional importance assessments of all business events that occur during the active period of business operations of an entity, and to classify them into importance levels and label them according to the assessment results; The neurodynamic replay module is used to compress and replay business events that meet the importance level requirements during the active business period using a compression replay algorithm during the dormant business period. The path optimization module is used to calculate the path optimization probability density of the transaction paths involved in each compressed replay of a business event based on the replay results of each compressed replay of the business event and the multi-dimensional importance assessment results of the business event. The adaptive path construction module is used to construct the transaction path network of the business entity based on the business entity's historical operating data and the historical supply chain data of all its partners, and to optimize the transaction path network based on the path optimization probability density of the transaction paths involved in each compressed replay of the business event. The visualization module is used to show the construction process of the transaction path network, the event distribution and importance level of each business event marked with an importance level, and the compressed replay process of the neurodynamic replay module.

[0092] Preferably, it also includes a parameter feedback optimization module, which is used to statistically analyze the adoption rate, transaction cost, and resource utilization rate of the best transaction path recommended by the transaction path network by users within a set period, and to dynamically adjust the replay enhancement coefficient β, the network connection strength coefficient γ of the transaction path, and the importance weight w_k of different business events in importance assessment under different dimensions.

[0093] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0094] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described above.

[0095] Example 2: The method for constructing the industrial internet transaction path optimization using a simulated neural network in this example is the same as in Example 1. The difference is that the transaction path network constructed using the method described in Example 1 can be used to recommend the best transaction path to the user.

[0096] Example 3: The method for constructing the industrial internet transaction path optimization using a simulated neural network in this example is the same as in Example 1. The difference is that it can also statistically analyze the adoption rate, transaction cost, and resource utilization rate of the best transaction path recommended by the transaction path network within a set period, and dynamically adjust the replay enhancement coefficient β, the network connection strength coefficient γ of the transaction path, and the importance weight w_k of different business events in importance assessment under different dimensions.

[0097] Based on all data from the past week or month, including the historical operating data of the business entity and the historical supply chain data of all its partners, as well as the accuracy (i.e., adoption rate) of the best transaction path recommended by the transaction path network, the replay enhancement coefficient β, the network connection strength coefficient γ of the transaction path, and the importance weight w_k of different business events in the importance assessment under different dimensions are dynamically adjusted.

[0098] Specifically, the steps of the gradient descent-based adaptive adjustment algorithm to dynamically adjust the replay enhancement coefficient β are as follows: First, define the loss function:

[0099] Where Accuracy(β) is the path accuracy when using β (the higher the value, the better), ConvergenceTime(β) is the time required for path optimization to converge (the shorter the value, the better), |β-β_default| is the regularization term to prevent β from deviating too far from the default value, and λ1 and λ2 are both trade-off coefficients; Calculate the gradient:

[0100] Gradient descent update to:

[0101] In the formula, η is the learning rate (typically 0.01-0.1). The constraint condition for β is: β_min and β_max are set values.

[0102] Assuming the current β=0.5, after one week of operation: Accuracy(0.5) = 0.85 ConvergenceTime(0.5) = 12 hours Estimate the gradient using the finite difference method:

[0103] Calculate the gradient of the loss function (assuming λ1=0.01, λ2=0.1, β_default=0.5):

[0104] Update β (assuming η = 0.1): β(t+1) = 0.5 - 0.1×(-0.13) = 0.5 + 0.013 = 0.513.

[0105] The steps for dynamically adjusting the network connectivity strength coefficient γ of a transaction path using a multi-objective optimization method based on network topology indicators are as follows: Define a multi-objective function: Objective1(γ) = NetworkConnectivity(γ) Objective2(γ) = NetworkRobustness(γ) Objective3(γ) = -ComputationalCost(γ) Pareto optimality solution: Use multi-objective optimization algorithms (such as NSGA-II) to find the optimal γ value on the Pareto front; Decision-makers' preferences: Based on the specific application scenario, select the γ that best meets the requirements on the Pareto front. 1. If connectivity is of greater importance: choose γ with a higher Objective1 value; 2. If robustness is of greater importance: choose γ with a higher Objective2 value; 3. If computing resources are limited: choose γ, which has a higher Objective-C (lower cost); Suppose that through multi-objective optimization, three candidate γ values ​​are found on the Pareto front: γ=0.3: Connectivity=0.75, Robustness=0.85, Cost=100 γ=0.5: Connectivity=0.90, Robustness=0.80, Cost=150 γ=0.7: Connectivity=0.95, Robustness=0.70, Cost=200 If the application scenario places greater emphasis on connectivity (such as the need to quickly discover new cooperation opportunities), choose γ=0.7; If robustness is of greater importance (such as in the financial industry, which requires a stable and reliable cooperative network), choose γ=0.3.

[0106] The steps of the reinforcement learning-based adaptive weight adjustment method to dynamically adjust the importance weights w_k of different business events in importance assessment under different dimensions are as follows: Modeling as a reinforcement learning problem: State: The current weight configuration s = (wk(1), wk(2), wk(3), wk(4)) Action: Adjust a weight a = Δw_k (increase or decrease w_k) Reward: r = α1·ΔAccuracy - α2·ΔCost + α3·ΔresourceUtilization In the formula, Δ represents the change relative to the previous period; Using the Q-learning algorithm: Q(s,a) ← Q(s,a) + η[r + γ·max_a' Q(s',a') - Q(s,a)] ε-greedy strategy: Explore randomly with probability ε (try new weight configurations) Utilize with probability (1-ε) (select the action with the highest current Q value); Weight normalization: After each adjustment, ensure that wk(1)+wk(2)+wk(3)+wk(4)=1.

[0107] Assume initial weights: w1=0.3, w2=0.3, w3=0.2, w4=0.2; Week 1: Try increasing wk(1) (transaction amount weight): wk(1)=0.35, wk(2)=0.28, wk(3)=0.19, wk(4)=0.18; Observations show that transaction costs decreased by 5% and resource utilization increased by 3%. The reward r = 0.5 × 0.05 - 0.3 × (-0.05) + 0.2 × 0.03 = 0.025 + 0.015 + 0.006 = 0.046 (positive reward); Update the Q value and note that this is a good direction for adjustment; Week 2: Continue to increase wk(1)=0.40, wk(2)=0.26, wk(3)=0.18, wk(4)=0.16; Observations: Transaction costs continued to decrease by 3%, but resource utilization began to decline by 1%. Reward r = 0.5 × 0.03 - 0.3 × (-0.03) + 0.2 × (-0.01) = 0.015 + 0.009 - 0.002 = 0.022 (positive reward but decreasing); The system learned that wk(1) should not be increased further. After multiple iterations, the optimal weights were converged: wk(1)=0.38, wk(2)=0.27, wk(3)=0.19, wk(4)=0.16.

[0108] It should be understood that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

Claims

1. A method for constructing an optimal transaction path for the industrial internet based on simulated neural networks, characterized in that, Specifically, the following steps are included: S1, construct the transaction path network of the operating entity based on the operating entity's historical operating data and the historical supply chain data of all its partners; S2 assesses the importance of all business events that occur during the active period of business operations of the business entity from multiple dimensions. S3, During the dormant period of business activities, the compression replay algorithm is used to compress and replay business events that meet the importance level requirements during the active period of business activities at a high frequency. S4. Based on the replay results of each compressed replay of a business event and the multi-dimensional importance assessment results of that business event, calculate the path optimization probability density of the transaction path involved in each compressed replay of a business event. S5, optimize the transaction path network based on the path optimization probability density of the transaction paths involved in each compressed replay of the business event; S6. Repeat steps S2-S5 to achieve closed-loop continuous optimization of the transaction path network.

2. The method for constructing an optimal industrial internet transaction path using a simulated neural network according to claim 1, characterized in that, In step S1, the specific steps for constructing the basic network of the business entity's transaction path based on the business entity's historical operating data and the historical supply chain data of all its partners are as follows: Collect historical operating data of business entities within a set time window, and obtain supply chain data of all their partners within the same time window through API interfaces; The business entity and its partners are treated as network nodes, and historical transaction relationships are treated as directed edges, forming a transaction path network for the business entity. The weight of each directed edge is initialized as a weighted value of historical transaction frequency and transaction amount.

3. The method for constructing an optimal industrial internet transaction path using a simulated neural network according to claim 1, characterized in that, In step S2, the specific steps for conducting a multi-dimensional importance assessment of all business events occurring during the active period of business operations are as follows: Real-time collection of operational data for all business events occurring during the active operating period of the business entity, as well as current supply chain data of relevant partners; The multidimensional importance score of each business event is calculated based on the multidimensional importance assessment function, and the importance level is assigned according to the score. Cache the operational data and importance assessment data for each operational event.

4. The method for constructing an optimal industrial internet transaction path using a simulated neural network according to claim 3, characterized in that, The formula for calculating the multidimensional importance assessment function is as follows: ， Where w_k is the importance weight of the kth business event in the mth dimension, f_k is the labeling function of the kth business event in the mth dimension, E_k(t) is the occurrence intensity of the kth business event in the mth dimension at time t, and m is the number of dimensions.

5. The method for constructing an optimal industrial internet transaction path using a simulated neural network according to claim 3, characterized in that, The specific steps in step S3, which use a compression replay algorithm to perform high-frequency compression replay of business events that meet the importance level requirements during the active period of a business entity's business activities, are as follows: Based on the importance level of all business events that occur during the active period of business operations, the business data of business events that meet the importance level requirements are extracted, and the extracted business events are used as replay events. Using a compression replay algorithm, the replay events are replayed at a set frequency with a time compression ratio of 10-100 times. During the replay process, the transaction paths involved in the replay events are recorded to obtain the event-path correlation.

6. The method for constructing an optimal industrial internet transaction path using a simulated neural network according to claim 5, characterized in that, The calculation formula for the compressed replay algorithm is as follows: ， Where K(t-τ) is the replay function, I(x, τ) is the importance score of the replay event, and H(τ) is the dormancy exponential function, where H(τ)=1 represents the dormancy period and H(τ)=0 represents the active period. .

7. The method for constructing an optimal industrial internet transaction path using a simulated neural network according to claim 5, characterized in that, In step S4, the specific steps for calculating the path optimization probability density of the transaction paths involved in the compressed replay of the business events are as follows: Calculate the network connection strength of a transaction path based on the event-path association of a certain replay event; The initiator of the transaction in the replay event is taken as the fore-synapse and the receiver as the after-synapse. The weighted comprehensive scoring method is used to calculate the intensity of the fore-synapse activity and the intensity of the after-synapse activity under the combined influence of the three dimensions of business activity, network connection, and decision response. The degree of change in synapse intensity is calculated based on the intensity of the fore-synapse activity and the intensity of the after-synapse activity. The path optimization probability density of the transaction path involved in the replay event is calculated based on the multi-dimensional importance score, compressed replay value, network connection strength and synaptic strength change degree of the transaction path involved in the replay event.

8. The method for constructing an optimal industrial internet transaction path using a simulated neural network according to claim 7, characterized in that, The formula for calculating the path optimization probability density of the transaction paths involved in the replay event is as follows: ， Wherein, P(x,t) is the path optimization probability density of the transaction path involved in the replay event, R(x,t) is the compressed replay value of the replay event, I(x,t) is the multi-dimensional importance score of the replay event, C(x,t) is the congestion degree of transaction traffic or data traffic at time t and transaction path x, α is the path diffusion coefficient of the transaction path involved in the replay event, β is the replay enhancement coefficient of the replay event, γ is the network connection strength coefficient of the transaction path involved in the replay event, and δ is the path attenuation coefficient of the transaction path involved in the replay event. The formula for calculating the network connection strength of the transaction path involved in the replay event is as follows: ， Where W0 is the basic connection strength of the transaction path involved in the replay event, d(i,j) is the distance between the initiating node and the receiving node in the replay event, λ is the connection attenuation length of the transaction path involved in the replay event, η is the transaction frequency enhancement factor between the initiating node and the receiving node in the replay event, and N(i,j) is the number of historical transactions between the initiating node and the receiving node in the replay event. The formula for calculating the degree of change in synaptic strength is: ， Where S(i,j,t) represents the degree of change in synaptic strength, A is the learning rate, τ is the synaptic decay time constant, pre(t) is the presynaptic activity intensity, and post(t-Δt) is the postsynaptic activity intensity. The formula for calculating the presynaptic activity intensity pre(t) is as follows: ， Wherein, l1(t) is the activity intensity of the presynapse at time t affected by the business activity dimension, l2(t) is the activity intensity of the presynapse at time t affected by the network connection dimension, l3(t) is the activity intensity of the presynapse at time t affected by the decision response dimension, W1 is the weight coefficient of the business activity dimension, W2 is the weight coefficient of the network connection dimension, and W3 is the weight coefficient of the decision response dimension. The formula for calculating the postsynaptic activity intensity post(t-Δt) is as follows: ， Where j1(t-Δt) is the activity intensity of the postsynapse at time t-Δt affected by the business activity dimension, j2(t-Δt) is the activity intensity of the postsynapse at time t-Δt affected by the network connection dimension, j3(t-Δt) is the activity intensity of the postsynapse at time t-Δt affected by the decision response dimension, V1 is the weight coefficient of the business activity dimension, V2 is the weight coefficient of the network connection dimension, V3 is the weight coefficient of the decision response dimension, and Δt is the time delay parameter.

9. The method for constructing an optimal industrial internet transaction path using a simulated neural network according to claim 1, characterized in that, In step S5, the specific steps for optimizing the transaction path network based on the path optimization probability density of the transaction paths involved in each compressed and replayed business event are as follows: Based on the path optimization probability density of the transaction paths involved in each compressed replay of an operational event, retain the transaction paths involved in operational events with a path optimization probability density greater than a set threshold, weaken the network connection strength of the transaction paths involved in operational events with a path optimization probability density less than a set threshold, or delete the transaction paths involved in operational events with a path optimization probability density less than a set threshold.

10. The method for constructing an optimal industrial internet transaction path using a simulated neural network according to claim 1, characterized in that, It also includes step S7, where the transaction path network recommends the best transaction path to the user.

11. The method for constructing an optimal industrial internet transaction path using a simulated neural network according to claim 10, characterized in that, It also includes step S8, which calculates the adoption rate, transaction cost, and resource utilization rate of the best transaction path recommended by the transaction path network for users within a set period, and dynamically adjusts the replay enhancement coefficient β, the network connection strength coefficient γ of the transaction path, and the importance weight w_k of different business events in the importance assessment under different dimensions.

12. A system for constructing an optimal transaction path for the industrial internet based on simulated neural networks, characterized in that: It includes modules for important event annotation, neurodynamic replay, path optimization, adaptive path construction, and visualization. The important event labeling module is used to conduct multi-dimensional importance assessments of all business events that occur during the active period of business operations of an entity, and to classify them into importance levels and label them according to the assessment results; The neurodynamic replay module is used to compress and replay business events that meet the importance level requirements during the active business period using a compression replay algorithm during the dormant business period. The path optimization module is used to calculate the path optimization probability density of the transaction paths involved in each compressed replay of a business event based on the replay results of each compressed replay of the business event and the multi-dimensional importance assessment results of the business event. The adaptive path construction module is used to construct the transaction path network of the business entity based on the business entity's historical operating data and the historical supply chain data of all its partners, and to optimize the transaction path network based on the path optimization probability density of the transaction paths involved in each compressed replay of the business event. The visualization module is used to show the construction process of the transaction path network, the event distribution and importance level of each business event marked with an importance level, and the compressed replay process of the neurodynamic replay module.

13. The industrial internet transaction path optimization and construction system based on simulated neural networks according to claim 12, characterized in that, It also includes a parameter feedback optimization module, which is used to statistically analyze the adoption rate, transaction cost, and resource utilization rate of the best transaction path recommended by the transaction path network within a set period, and to dynamically adjust the replay enhancement coefficient β, the network connection strength coefficient γ of the transaction path, and the importance weight w_k of different business events in the importance assessment under different dimensions.

14. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-11.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-11.

Images