Display voice interaction system and method

Abstract

The invention relates to the technical field of voice interaction, and discloses a display voice interaction system and method.The display voice interaction method comprises the following steps that representation data is quantified through the distributed game theory and the calculation safety performance relation, and a resource optimal allocation strategy matrix is generated; generating a weight coefficient set through the data of the secure calculation entropy model, the resource optimal allocation strategy matrix and the three-factor balance model; through the weight coefficient set, the dual-channel parallel processing system generates a dual-channel parallel execution strategy table and a result release strategy table; generating a distributed elastic calculation fault-tolerant scheme data packet through a double-channel parallel execution strategy table and a result release strategy table; and generating an adaptive security configuration data set through security calculation entropy model data, a dual-channel parallel execution strategy table, a result release strategy table and a distributed elastic calculation fault-tolerant scheme data packet.

Description

Technical Field

[0001] The present invention relates to the technical field of voice interaction, and more specifically, it relates to a display voice interaction system and method. Background Art

[0002] With the continuous development of human-computer interaction technology, voice interaction has become an important interaction method for display devices and is widely used in many fields such as enterprise office, financial transactions, medical information systems, and smart homes. Existing display voice interaction systems mainly adopt cloud processing architectures, local processing architectures, and hybrid processing architectures. Although the cloud processing architecture can utilize the powerful computing resources of the cloud for complex voice recognition and security verification, it has strong network dependence and high response latency; the local processing architecture has a fast response speed but is limited by the computing power of the terminal and has great security risks; the hybrid processing architecture also fails to fundamentally solve the contradiction between response speed and security. Moreover, the existing technologies generally adopt a serial processing mode and have many problems:

[0003] The inherent contradiction between security and response speed: To ensure a fast response, the security verification process is often simplified, resulting in security risks; strengthening security verification significantly increases the response latency and affects the user experience; limitations of the verification mechanism: Some systems only perform a one-time identity verification at the beginning of the session, with a risk of session hijacking; the continuous verification mechanism increases the system load and response latency; unreasonable resource allocation: Adopting a static resource allocation method, it is impossible to dynamically adjust resource allocation according to the security sensitivity of instructions and system load, and the resource utilization efficiency is low; lack of elasticity and fault tolerance: In the event of node failures or network fluctuations, there is no effective task migration and load balancing mechanism, which easily leads to service interruptions or performance degradation.

[0004] Therefore, there is an urgent need to develop a new display voice interaction system that can break through the inherent contradiction of "response speed and depth of security verification", realizing continuous and robust security verification while ensuring an extremely low interaction latency (100 ms), and having elasticity and fault tolerance in the case of unstable networks and partial node failures. Summary of the Invention

[0005] The present invention provides a display voice interaction method, including:

[0006] Quantifying and characterizing data through distributed game theory and computational security performance relationships to generate an optimal resource allocation strategy matrix;

[0007] Generating a weight coefficient set through the secure computational entropy model data, the optimal resource allocation strategy matrix, and the three-factor balance model;

[0008] Generating a dual-channel parallel execution strategy table and a result release strategy table through the weight coefficient set and the dual-channel parallel processing system;

[0009] Generate a distributed elastic computing fault tolerance solution data packet through a dual-channel parallel execution policy table and a result release policy table;

[0010] Generate an adaptive security configuration data set through secure computing entropy model data, a dual-channel parallel execution policy table, a result release policy table, and a distributed elastic computing fault tolerance solution data packet.

[0011] In a preferred embodiment, in the step of constructing the secure computing entropy model, the secure computing entropy model includes: an entropy calculation core unit, a correlation weight matrix unit, and an entropy balance constraint unit; the entropy balance constraint unit applies an entropy balance constraint algorithm to the secure computing entropy value to ensure that the secure computing entropy value does not exceed the theoretical threshold, and determines the theoretical threshold of the secure computing entropy value using the sizes of the performance metric set and the security metric set, so as to ensure that the system operates within a reasonable range and maintain the balance between security and computing efficiency.

[0012] In a preferred embodiment, in the step of generating an optimal resource allocation policy matrix by distributed game theory, the distributed non-zero-sum game model consists of an agent model unit, a utility function calculation unit, a Nash equilibrium solving unit, and a policy optimization unit to form a utility function. The calculation formula for the utility function value of each agent in the agent model unit is as follows:

[0013]

[0014] where, U i represents the utility function value of agent Player i ; a i represents the strategy of agent Player i ; represents the set of strategies of other agents except agent Player i ; α i is a coefficient related to the performance of agent Player i ; is the performance function value of agent Player i when taking strategy a i and other agents take the set of strategies a -i ; β i is a coefficient related to the security of agent Player i ; is the security function value of agent Player i when taking strategy a i and other agents take the set of strategies a1.

[0015] In a preferred embodiment, in the step of the three-factor balance model generating the weight coefficient set, the three-factor balance model consists of an efficiency evaluation unit, a collaborative factor calculation unit, and a weight optimization unit. For the efficiency evaluation unit, the calculation formula for the overall system efficiency value is as follows:

[0016] E(C, S) = α(t)·P c (C) + β(t)·P s (S) + γ(t)·Syn(C, S);

[0017] Wherein, E(C, S) represents the overall system efficiency value; C is the computing resource allocation vector; S is the security resource allocation vector; α(t), β(t), and γ(t) are time-varying weight coefficients. α(t) is used to measure the weight of the computing performance function P c (C) in the calculation of the overall system efficiency; β(t) is used to measure the weight of the security performance function P s (S) in the calculation of the overall system efficiency; γ(t) is used to measure the weight of the collaborative factor Syn(C, S) in the calculation of the overall system efficiency; P c (C) is the computing performance function related to the computing resource allocation vector C; P s (S) is the security performance function related to the security resource allocation vector S; Syn(C, S) is the collaborative factor.

[0018] In a preferred embodiment, the calculation formula for the collaborative factor is as follows:

[0019]

[0020] Wherein, Synb(C, S) represents the collaborative factor value; C is the computing resource allocation vector, and C i is the i-th element in the computing resource allocation vector C, representing the allocation amount of the i-th computing resource; S is the security resource allocation vector, and S j is the j-th element in the security resource allocation vector S; n is the number of elements in the computing resource allocation vector C; m is the number of elements in the security resource allocation vector S; δ ij is the collaborative coefficient between the computing resource C i and the security resource S j ; (1 - (C i - S j )) is used to adjust the influence of the difference in the allocation amounts of the computing resource and the security resource on the collaborative factor value.

[0021] In a preferred embodiment, in the steps of the dual-channel parallel processing system generating the dual-channel parallel execution policy table and the result release policy table, the dual-channel parallel processing system is composed of an instruction feature extraction unit, a security sensitivity calculation unit, a policy generation unit, and a result release unit. The formula for calculating the security sensitivity value is as follows:

[0022]

[0023] Wherein, S(V i ) represents the security sensitivity value of the instruction; V i represents the instruction feature vector; k is the number of elements in the instruction feature vector V i ; w j is the feature weight coefficient; v j is the j-th element in the instruction feature vector V i , representing the value of the j-th instruction feature; R j is the basic risk coefficient. By multiplying and accumulating these parameters, the security sensitivity value of the instruction is obtained.

[0024] In a preferred embodiment, the policy generation unit applies the decision function Strategy(V i ) to generate dual-channel processing policies, including a full parallel policy, a partial parallel policy, and a verification priority policy, and selects the corresponding policy according to the comparison result of the instruction security sensitivity value and the preset threshold.

[0025] In a preferred embodiment, in the steps of generating the distributed elastic computing fault tolerance scheme data packet, the distributed elastic computing fault tolerance system is composed of a node health monitoring unit, a status classification unit, a task matching degree calculation unit, a task allocation unit, and a task migration unit. The formula for the health index is as follows:

[0026] H i =ω1·CPU i +ω2·MEM i +ω3·NET i +ω4·RESP i ;

[0027] Wherein, H i represents the health index of the i-th computing node; ω1, ω2, ω3, and ω4 are the weight coefficients corresponding to CPU i , MEM i , NET i , and RESP i respectively; CPU i represents the normalized CPU load of the node; MEM i represents the normalized memory usage rate of the node; NET iRepresents the normalized value of the network connection quality of a node. The normalized network connection quality data is used to reflect the network status in the calculation of the health index; RESP i Represents the normalized value of the response time of a node.

[0028] In a preferred embodiment, the task allocation unit applies an optimal task allocation algorithm to generate a task allocation scheme. The calculation formula of its optimization objective function is as follows:

[0029]

[0030] Where, A * Represents the optimal task allocation scheme; A is the set of all possible task allocation schemes, and argmax A Represents finding the task allocation scheme in set A that maximizes the value of the subsequent summation expression; m is the number of tasks, and n is the number of nodes; A ji Represents whether task T j is assigned to node N i ; The variable of j , N i ; Match(T j , N i ) is the matching degree function between task T j and node N i , which is used to evaluate the adaptation degree between task T ji and node N j , i . By multiplying A

[0031] In a preferred embodiment, a display voice interaction system includes:

[0032] A policy matrix module that quantifies and represents data through distributed game theory and computational security performance relationships, and generates an optimal resource allocation policy matrix;

[0033] A weight coefficient module that generates a set of weight coefficients through a three-factor balance model using secure computational entropy model data and the optimal resource allocation policy matrix;

[0034] A processing policy table module that generates a dual-channel parallel execution policy table and a result release policy table through a dual-channel parallel processing system using the set of weight coefficients;

[0035] A scheme data module that generates a distributed elastic computing fault tolerance scheme data packet through the dual-channel parallel execution policy table and the result release policy table;

[0036] ​​The security configuration module generates an adaptive security configuration data set by means of secure computing entropy model data, a dual-channel parallel execution policy table, a result release policy table, and a distributed elastic computing fault tolerance solution data packet.

[0037] The beneficial effects of the present invention are as follows:

[0038] Balance between response latency and security: A breakthrough balance between response latency and security is achieved. Under high security requirements, the response latency can be controlled within 100 ms, solving the problem in the prior art that it is difficult to balance the two.

[0039] Improved resource utilization efficiency: The system resource utilization efficiency is optimized. Through a dynamic resource allocation mechanism, resources are allocated according to the security sensitivity of different instructions and the system load, improving resource utilization rate and overall performance.

[0040] Enhanced elasticity and fault tolerance: The system has extremely strong system elasticity and fault tolerance. In the case of partial node failures, it can still maintain a high level of functional integrity and performance, ensuring service continuity.

[0041] Dynamic allocation of security resources: Intelligent dynamic allocation of security resources is realized, reducing computing resource consumption and improving resource utilization efficiency while ensuring security.

[0042] Reduced code complexity and maintenance cost: Through innovative architecture and algorithm design, the code complexity is reduced, thereby reducing the maintenance cost.

[0043] Excellent scalability and versatility: The system has excellent scalability and versatility and can adapt to changes in different scenarios and requirements.

[0044] Improved user experience: The innovative adaptive security boundary improves the user experience, providing users with a more stable, efficient, and secure interaction environment.

[0045] Energy consumption optimization: Substantial energy consumption optimization is achieved, reducing energy consumption while ensuring system performance. Description of the Drawings

[0046] Figure 1 It is a flowchart of a method for voice interaction of a display of the present invention. Detailed Embodiments

[0047] Reference will now be made to the exemplary embodiments to discuss the subject matter described herein. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and thereby implement the subject matter described herein, and that changes can be made to the functions and arrangements of the elements discussed without departing from the scope of protection of the content of this specification. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0048] In at least one embodiment of the present invention, a method for voice interaction of a display is disclosed. By using distributed game theory and computational security performance relationship to quantify and characterize data, a resource optimal allocation strategy matrix is generated, as Figure 1 shown, including the following steps:

[0049] Step100, construct a secure computing entropy model to generate computational security performance relationship quantification and characterization data;

[0050] Input system performance index data and security index data, and generate secure computing entropy model data through a secure computing entropy modeling algorithm. The secure computing entropy model consists of an entropy calculation core unit, a correlation weight matrix unit, and an entropy balance constraint unit.

[0051] Step101, the system collects two types of index data as the input of the secure computing entropy model: the system performance index set, and the calculation formula is as follows:

[0052] P = p1, p2,..., p n ;

[0053] where, P represents the system performance index set; n represents the number of elements in the system performance index set; p i represents the normalized value of the i-th performance index.

[0054] The calculation formula of the security index set is as follows:

[0055] S = s1, s2,..., s m ;

[0056] where, S represents the security index set; m represents the number of elements in the security index set; s j represents the normalized value of the j-th security index.

[0057] Step102, the entropy calculation core unit receives the performance index set P and the security index set S as inputs, and calculates the secure computing entropy value H(P, S), and the calculation formula is as follows:

[0058]

[0059] Among them, H(P, S) represents the security calculation entropy value of the system; n is the number of elements in the system performance index set P = {p1, p2,..., p n}; m is the number of elements in the security index set S = {s1, s2,..., s m}; ω ij is the element in the associated weight matrix unit; p i is the normalized value of the i-th performance index in the system performance index set; s j is the normalized value of the j-th security index in the security index set; log() is the logarithmic function with the natural logarithm as the base.

[0060] Step103, output the security calculation entropy model data, including: the security calculation entropy value H(P, S), and the calculation formula of the associated weight matrix is as follows:

[0061] W = (ω ij ) n×m ;

[0062] Among them, W is the associated weight matrix, and ω ij represents the associated weight between the performance index p i and the security index s j . n is the number of elements in the system performance index set P, and m is the number of elements in the security index set S.

[0063] The entropy balance constraint parameters are as follows:

[0064] H max , n, m;

[0065] Among them, H max represents the theoretical maximum entropy value, n is the number of elements in the system performance index set P, and P represents the system performance index set; m is the number of elements in the security index set S, and S represents the security index set.

[0066] Step200, generate the optimal resource allocation strategy matrix based on the distributed game theory;

[0067] The optimal allocation strategy matrix is used to guide the system effectiveness balance optimization and dual-channel parallel processing. The optimal allocation strategy matrix is generated by the distributed non-zero-sum game model, which consists of an agent model unit, a utility function calculation unit, a Nash equilibrium solving unit, and a strategy optimization unit. The agent model unit maps the computing nodes in the distributed edge security symbiotic display voice interaction network to game agents; the utility function calculation unit calculates the utility function value for each agent; the Nash equilibrium solving unit finds the strategy combination that meets specific conditions; the strategy optimization unit approximates the Nash equilibrium point through the gradient ascent iteration algorithm, and finally obtains the optimal resource allocation strategy matrix to achieve the optimal allocation of resources.

[0068] In at least one embodiment of the present invention, the secure computing entropy model data in Input Step 100 is used to generate an optimal resource allocation strategy matrix through a distributed non-zero-sum game model. The distributed non-zero-sum game model consists of an agent model unit, a utility function calculation unit, a Nash equilibrium solving unit, and a strategy optimization unit.

[0069] Step 201, the agent model unit receives the system topology data and maps the computing nodes in the distributed edge security symbiotic display voice interaction network into game agents;

[0070] Step 202, the utility function calculation unit takes the secure computing entropy model data, the agent model, and the strategy combination as inputs, and calculates the utility function value for each agent. The calculation formula is as follows:

[0071]

[0072] Where U i represents the utility function value of agent Player i ; a i represents the strategy of agent Player i ; represents the set of strategies of other agents except agent Player i ; α i is a coefficient related to the performance of agent Player i ; is the performance function value of agent Player i when taking strategy a i and other agents take the set of strategies ; β i is a coefficient related to the security of agent Player i ; is the security function value of agent Player i when taking strategy a i and other agents take the set of strategies ;

[0073] Step 203, the Nash equilibrium solving unit receives the output of the utility function calculation unit and applies the dynamic Nash equilibrium algorithm to find a strategy combination that satisfies the following conditions. The calculation formula is as follows:

[0074]

[0075] Where represents agent Player i when all other agents take the optimal set of strategies ; represents agent Playeri when all other agents adopt the optimal strategy set ; denotes for agent Player i in the strategy space A i any strategy a i ; denotes for any agent Player in the agent set N = 1, 2,..., n i .

[0076] Step204, the strategy optimization unit takes the initial strategy combination and the Nash equilibrium condition as inputs, and approximates the Nash equilibrium point through the gradient ascent iteration algorithm. The calculation formula is as follows:

[0077]

[0078] where denotes the strategy of agent Player i at the t-th iteration. denotes the strategy of agent Player i at the (t + 1)-th iteration. η represents the learning rate parameter, which controls the step size of strategy update. denotes the gradient of the utility function with respect to strategy a i .

[0079] Step205, output the optimal resource allocation strategy matrix. The calculation formula is as follows:

[0080]

[0081] where A * denotes the optimal resource allocation strategy matrix. For the element its row is the i-th row. Since is a vector, assuming the vector dimension is k, then its column is from column 1 to column k. is a vector, denoting the optimal resource allocation strategy of agent Player i ; n represents the number of agents, and the value range of i is from 1 to n, corresponding to different agents.

[0082] Step300, through the secure calculation entropy model data and the optimal resource allocation strategy matrix, the three-factor balance model generates a weight coefficient set;

[0083] The generation of the weight coefficient set by the three-factor balance model means that, taking the security calculation entropy model data of Step100 and the resource optimal allocation strategy matrix of Step200 as inputs, a weight coefficient set for balancing computing performance, security performance, and their synergy effect is generated through the three-factor balance model composed of an efficiency evaluation unit, a synergy factor calculation unit, and a weight optimization unit.

[0084] Step301, the efficiency evaluation unit receives the resource allocation data and calculates the overall system efficiency value. The calculation formula is as follows:

[0085] E(C, S) = α(t)·P c (C) + β(t)·P s (S) + γ(t)·Syn(C, S);

[0086] Among them, E(C, S) represents the overall system efficiency value; C is the computing resource allocation vector; S is the security resource allocation vector; α(t), β(t), and γ(t) are time-varying weight coefficients. α(t) is used to measure the weight of the computing performance function P c (C) in the calculation of the overall system efficiency; β(t) is used to measure the weight of the security performance function P s (S) in the calculation of the overall system efficiency; γ(t) is used to measure the weight of the synergy factor Svn(C, S) in the calculation of the overall system efficiency; P c (C) is the computing performance function related to the computing resource allocation vector C; P s (S) is the security performance function related to the security resource allocation vector S; Syn(C, S) is the synergy factor.

[0087] Step302, the synergy factor calculation unit receives the computing resource allocation vector C and the security resource allocation vector S. The calculation formula for the synergy factor value is as follows:

[0088]

[0089] Among them, Syn(C, S) represents the synergy factor value; C is the computing resource allocation vector, C i is the i-th element in the computing resource allocation vector C; S is the security resource allocation vector, S j is the j-th element in the security resource allocation vector S; n is the number of elements in the computing resource allocation vector C; m is the number of elements in the security resource allocation vector S; δ ij is the synergy coefficient between the computing resource C i and the security resource S j ; (1 - (C i - S j)) It is used to adjust the impact of the difference in the allocation amounts of computing resources and security resources on the collaborative factor value.

[0090] Step303, the weight optimization unit receives the outputs of the performance evaluation unit and the collaborative factor calculation unit, and dynamically adjusts the weight coefficients through the three-factor weight optimization algorithm. The calculation formula is as follows:

[0091]

[0092] Among them, α(t), β(t), and γ(t) respectively represent the computing performance weight, security performance weight, and collaborative factor weight at the t-th iteration; α(t + 1), β(t + 1), and γ(t + 1) respectively represent the three weight coefficients at the (t + 1)-th iteration; λ represents the learning rate parameter; represents the gradient vector of the performance function E(C, S) with respect to the weight coefficients α, β, γ

[0093] Step304, through iterative optimization, output the three-factor balanced weight coefficient set. The calculation formula is as follows:

[0094] W * = α * (t), β * (t), γ * (t);

[0095] Among them, W * represents the three-factor balanced weight coefficient set. α * (t) represents the optimal computing performance weight at time t. β * (t) represents the optimal security performance weight at time t. γ * (t) represents the optimal collaborative factor weight at time t. The three-factor balanced weight coefficient set is used as the input of Step400 to guide the selection of processing strategies and the formulation of result release strategies in the dual-channel parallel architecture.

[0096] Step400, through the weight coefficient set, the dual-channel parallel processing system generates a dual-channel parallel execution strategy table and a result release strategy table;

[0097] The dual-channel parallel execution strategy table is a table that determines which processing strategy (fully parallel strategy, partially parallel strategy, or verification priority strategy) to adopt for each voice instruction according to the security sensitivity value of the instruction.

[0098] Definition of the result release strategy table: According to the processing strategy determined in the dual-channel parallel execution strategy table, it stipulates how the calculation results are released (for example, in the full parallel strategy, the calculation processing and security verification are carried out simultaneously without interference, and the results can be directly released; in the partial parallel strategy, the calculation results need to be released in grades according to the verification progress; in the verification priority strategy, the calculation processing is only executed after the security verification is completed and passed, and the results are released after the verification passes).

[0099] Input the three-factor balance weight coefficient set of Step300, and generate the dual-channel parallel execution strategy table and the result release strategy table through the dual-channel parallel processing system. The dual-channel parallel processing system consists of an instruction feature extraction unit, a security sensitivity calculation unit, a strategy generation unit, and a result release unit.

[0100] Step401, the instruction feature extraction unit receives the voice instruction data, performs feature extraction and basic semantic understanding, and outputs the instruction feature vector. The calculation formula is as follows:

[0101] V i =v1, v2,..., v k ;

[0102] Among them, V i represents the feature vector of the i-th voice instruction; v j (j = 1, 2,..., k) represents the j-th component of the feature vector; k represents the feature dimension, usually 1030, depending on the application scenario.

[0103] The instruction feature extraction unit adopts a lightweight semantic parsing algorithm to complete feature extraction within 50ms, laying a foundation for subsequent processing.

[0104] Step402, the security sensitivity calculation unit receives the instruction feature vector and calculates the security sensitivity value of the instruction. The calculation formula is as follows:

[0105]

[0106] Among them, S(V i ) represents the security sensitivity value of the instruction V i ; w j represents the weight coefficient of the j-th feature; v j represents the j-th component of the feature vector V i ; R j represents the basic risk coefficient of the j-th feature; k represents the feature dimension, usually 1030, depending on the application scenario.

[0107] Step403, the strategy generation unit receives the security sensitivity value and generates the dual-channel processing strategy. Its decision function is as follows, and the calculation formula is as follows:

[0108]

[0109] Among them, Strategy(V i ) represents the processing strategy adopted for the instruction V i ; V i represents the feature vector of the i-th voice instruction; S(V i ) represents the security sensitivity value of the instruction V i ; T low represents the low-risk threshold; T high represents the high-risk threshold. FullParallel represents the full parallel strategy; PartialParallel represents the partial parallel strategy; VerifyFirst represents the verification-first strategy.

[0110] Step404, the result release unit receives the output of the strategy generation unit and the real-time verification progress data, and generates a result release strategy. The calculation formula of its decision function is as follows:

[0111]

[0112] Among them, Release(R c , P v ) represents the release strategy adopted for the calculation result R v according to the verification progress P c ; FykkRekease represents the full release strategy; PartialRelease represents the partial release strategy; HoldRelease represents the hold release strategy; P v represents the progress percentage of security verification; T p represents the partial release threshold; T f represents the full release threshold; f(P v ) represents the result filtering function based on the verification progress.

[0113] Step405, output the following data: The calculation formula of the dual-channel parallel execution strategy table is as follows:

[0114] ST = (V i , Strategy(V i ))|i = 1, 2,..., n;

[0115] Among them, ST represents the dual-channel parallel execution strategy table. V i represents the feature vector of the i-th voice instruction, and the value range of i is from 1 to n; n represents the total number of voice instructions; Strategy(V i ) represents the processing strategy adopted for the instruction V i .

[0116] The calculation formula of the result release policy table is as follows:

[0117] RT = (P v , Release(R c , P v )) | P v ∈(0, 1);

[0118] Among them, RT represents the result release policy table, and P v represents the progress percentage of security verification. Release(R c , P v ) represents the release policy adopted for the calculation result R v according to the verification progress P c .

[0119] The dual-channel parallel execution policy table and the result release policy table are used as the inputs of Step500 and Step600 to guide the implementation of the distributed elastic computing fault tolerance mechanism and the dynamic adjustment of the security boundary.

[0120] Step500, through the dual-channel parallel execution policy table and the result release policy table, generates a distributed elastic computing fault tolerance solution data packet, including:

[0121] The node health monitoring parameter set is as follows:

[0122] HM = ω1, ω2, ω3, ω4;

[0123] Among them, ω1, ω2, ω3, w4 are parameters defining the calculation method of the node health index.

[0124] Define the calculation method of the node health index, and the state classification threshold set is as follows

[0125] SC = T h , T c ;

[0126] Among them, T h represents the health threshold. T c represents the critical threshold.

[0127] Define the node state classification standard, and the task matching degree parameter set is as follows:

[0128] MP = α, β, γ;

[0129] Among them, α, β, and γ are the weight coefficients of three factors.

[0130] Define the priority consideration factors for task allocation, and the optimal task allocation matrix is as follows:

[0131] A * = (A ji )m×n ;

[0132] Among them, A * represents the optimal task allocation scheme. A ji represents a 0-1 variable. m represents the total number of tasks in the system. n represents the total number of nodes in the system.

[0133] Define the optimal task allocation scheme and the task migration plan set in the current state as follows:

[0134] TP = {Migrate(T j , N src , N dst ) | T j needs to be migrated};

[0135] Migrate(T j , N src , N dst ) represents the operation set of migrating task T j from the source node N s rc to the target node N d st. StateSync(T j ) represents the task status synchronization operation.

[0136] DataTransfer(T j ) represents the data transfer operation; ProgressRestotre(T j ) represents the progress restoration operation.

[0137] The distributed elastic computing fault tolerance scheme data packet refers to the data set generated by the distributed elastic computing fault tolerance system based on the dual-channel parallel execution strategy table and the result release strategy table output by Step400. This data packet contains the key information for implementing the distributed elastic computing fault tolerance mechanism, guiding the system on how to reasonably allocate and migrate tasks in the face of abnormal situations such as node failures, so as to ensure the normal operation and stable performance of the system.

[0138] Step501, the node health monitoring unit receives the system node status data and calculates the health index of each computing node. The calculation formula is as follows:

[0139] H i = w1·CPU i + w2·MEM i + w3·NET i + w4·RESP i ;

[0140] Among them, H i represents the health index of the i-th computing node; ω1, ω2, ω3, and ω4 are respectively related to CPUi , MEM i , NET i and RESP i corresponding weight coefficients; CPU i represents the normalized CPU load value of the node; MEM i represents the normalized memory usage rate of the node; NET i represents the normalized network connection quality value of the node; RESP i represents the normalized response time value of the node.

[0141] Step502, the status classification unit receives the node health index and classifies the nodes into three categories by applying the node status classification algorithm. The calculation formula of its classification function is as follows:

[0142]

[0143] where Status(N i ) represents the status category of node N i ; Healthy represents the healthy state; Degraded represents the performance degradation state, that is, the node can undertake part of the computing tasks; Critical represents the critical state; H i represents the health index of the i-th computing node,

[0144] Step503, the task matching degree calculation unit receives the node status data and the task characteristic data, and calculates the matching degree matrix between the task and the node. The calculation formula is as follows:

[0145]

[0146] where Match(T j , N i ) represents the matching degree between task T j and node N i ; α, β, and γ represent the weight coefficients of the three factors, and satisfy α + β + γ = 1; Capability(N i , T j ) represents the node processing ability matching function; Proximity(N i , T j ) represents the data proximity function; LoadBalance(N i ) represents the load balancing function.

[0147] Step504, the task allocation unit receives the matching degree matrix and generates a task allocation plan by applying the optimal task allocation algorithm. The calculation formula of its optimization objective function is as follows:

[0148]

[0149] Among them, A * represents the optimal task allocation scheme; A is the set of all possible task allocation schemes, and argmax A represents finding the task allocation scheme in set A that maximizes the value of the subsequent summation expression; m is the number of tasks, and n is the number of nodes; A ji represents whether task T j is allocated to node N i is a variable; Mathch(T j , N i ) is the matching degree function between task T j and node N i .

[0150] Step505, the task migration unit receives the current task allocation status and the optimal task allocation scheme, and generates a task migration execution plan. The calculation formula is as follows:

[0151]

[0152] Among them, Migrate(T j , N src , N dst ) represents the set of operations to migrate task T j from the source node N src to the target node N dst ; StateSync(T j ) represents the task status synchronization operation; DataTransfer(T j ) represents the data transfer operation; ProgressRestore(T j ) represents the progress restoration operation.

[0153] Step600, generate an adaptive security configuration data set by securely calculating entropy model data, a dual-channel parallel execution policy table, a result release policy table, and a distributed elastic computing fault tolerance scheme data packet;

[0154] Generating an adaptive security configuration data set means inputting the security calculation entropy model data in Step100, the dual-channel parallel execution policy table and the result release policy table in Step400, and the distributed elastic computing fault tolerance scheme data packet in Step500, and generating an adaptive security configuration data set through an adaptive security boundary system composed of a risk factor analysis unit, a risk level assessment unit, a security verification configuration unit, a resource allocation unit, and a policy smooth transition unit.

[0155] Step601, the risk factor analysis unit receives system operation environment data and user interaction data, and extracts the risk factor value. The calculation formula is as follows:

[0156] RF = rf1, rf2, ..., rf l ;

[0157] Among them, RF represents the set of system risk factors; rf i (i = 1, 2, ..., l) represents the quantification value of the i-th risk factor; (rf1) is the abnormal degree of user behavior; (rf2) is the operation sensitivity; (rf3) is the environmental threat level; (rf4) is the historical security event frequency; (rf5) is the network environment credibility, etc.

[0158] Step602, the risk level assessment unit receives the set of risk factors and calculates the current system risk level by applying the multi-factor risk assessment algorithm. The calculation formula is as follows:

[0159]

[0160] Among them, RL represents the current system risk level; φ i represents the weight coefficient of the i-th risk factor, reflecting the relative importance of this factor, and satisfies rf i represents the quantification value of the i-th risk factor; I i represents the influence coefficient of the i-th risk factor.

[0161] Step603, the security verification configuration unit receives the system risk level and generates security verification configuration data, including:

[0162] The security verification intensity value is calculated by the verification intensity mapping function as follows:

[0163] SV = f sv (RL) = SV base +ΔSV·sigmoid(k·RL - RL mid ));

[0164] Among them, SV represents the security verification intensity value; f sv (RL) is the verification intensity mapping function regarding the current system risk level RL; SV base represents the basic verification intensity; ΔSV represents the maximum adjustment amount of the verification intensity; sigmoid() represents the S-shaped function; k represents the adjustment sensitivity parameter; RL represents the current system risk level; RL mid represents the intermediate risk level.

[0165] The set of verification method combinations is determined by the verification method selection function. The calculation formula is as follows:

[0166] VM = f vm (RL) = vm1, vm2, ..., vm q|P(vm i |RL)>T vm ;

[0167] Wherein, VM represents a selected set of verification method combinations; f vm (RL) is a verification method selection function regarding the current risk level RL of the system; vm i represents the i-th verification method; P(vm i |RL) represents the probability of selecting the verification method vm at the current risk level RL i ; T vm represents the method selection threshold.

[0168] Step604, the resource allocation unit receives the system risk level and security verification configuration data, and generates a security resource allocation plan. The calculation formula is as follows:

[0169]

[0170] Wherein, SR i represents the amount of resources allocated to the i-th type of security function; SR base，i represents the basic resource allocation amount of the i-th type of security function; ΔSR i represents the maximum additional resource amount of the i-th type of security function; a i represents the resource allocation coefficient of the i-th type of security function; RL represents the current risk level of the system; r represents the total number of categories of security functions; represents calculating the sum of for all security function categories j from 1 to r; This part represents the i-th type of security function considering the risk level RL and the resource allocation coefficient a i case; represents the additional resource allocation amount of the i-th type of security function based on the risk level RL; Then the finally obtained amount of resources SR allocated to the i-th type of security function i .

[0171] Step605, the policy smooth transition unit receives the current security policy state and the target security policy state, and applies the smooth transition algorithm to generate a transition security policy. The calculation formula is as follows:

[0172] SP(t + 1) = (1 - λ)·SP(t) + λ·SP target ;

[0173] Wherein, SP(t) represents the security policy state at time t; SP(t + 1) represents the security policy state at time t + 1; SP target represents the target security policy state; λ represents the smooth transition coefficient.

[0174] Step606, output the adaptive security configuration data set, including: The calculation formula of the risk assessment parameter set is as follows:

[0175] RP = φ1, φ2, …, φ l , I1, I2, …, I l ;

[0176] Among them, φ1, φ2, …, φ1 represent the weights corresponding to different risk factors. φ i represents the weight coefficient of the i-th risk factor. I i represents the influence coefficient of the i-th risk factor. This risk assessment parameter set defines the weights and influence coefficients of risk factors. The greater the weight, the greater the impact of this risk factor on the final risk assessment result.

[0177] I1, I2, I l respectively represent the influence coefficients of the 1st, 2nd, l-th risk factors, reflecting the specific impact degree of each risk factor on the system risk state. Assume that in this system, I1 represents the influence coefficient of network connection stability on risk, and its value range is [0, 1]. When the network connection is stable, I1 takes the value of 0.2; when the network connection is unstable, I1 takes the value of 0.8. I2 represents the influence coefficient of user operation frequency on risk, and its value range is [0, 1]. If the user operation frequency is low, I2 takes the value of 0.3; if the user operation frequency is high, I2 takes the value of 0.7. These coefficients and weights act together to define the weights and influence coefficients of risk factors to accurately evaluate the risks faced by the system. The verification strength configuration parameter set, the calculation formula is as follows:

[0178] VC = SV base , ΔSV, k, RL mid ;

[0179] Among them, SV base represents the basic verification strength. ΔSV represents the adjustment increment of the verification strength. k is a coefficient related to the verification strength adjustment. RL mid represents the median value of the risk level.

[0180] The verification method combination mapping table, the calculation formula is as follows:

[0181] VM table =(RL i , VM i )|i = 1, 2, …, z;

[0182] Among them, RL i represents the i-th risk level. VM i represents the corresponding to the risk level RL iThe selected combination of verification methods, i.e., when the system is at risk level RL i it will adopt a series of verification methods included in VM i to define the selection of verification methods under different risk levels.

[0183] The resource allocation plan data has the following calculation formula:

[0184] RA = SR base，i , ΔSR i , a i |i = 1, 2,..., r;

[0185] where SR base，i represents the basic resource allocation amount of the i-th type of security function. ΔSR i represents the maximum additional resource amount of the i-th type of security function. a i represents the resource allocation coefficient of the i-th type of security function; λ is the policy smooth transition parameter.

[0186] The adaptive security configuration data set, together with the output of the previous steps, constitutes the technical implementation basis of the complete distributed edge security symbiotic display voice interaction network, enabling the system to maintain high security while ensuring low-latency response, and having environmental adaptability and fault recovery capabilities.

[0187] The deployment of this embodiment in the intelligent display voice interaction system of the financial trading hall has achieved two of the most important technical effects: a breakthrough balance between response latency and security, and a significant improvement in system elasticity and fault tolerance. The following are the verification data for these two aspects of technical effects.

[0188] Through the data collection and analysis of 238 display terminals in the trading hall during a month's trading days (22 working days), the performance comparison between this embodiment and the traditional serial processing technical solution is as follows:

[0189] The performance comparison is shown in the following table:

[0190]

[0191] The data shows that this embodiment has achieved a significant reduction in response latency in all operations at all security sensitivity levels. Especially for high-sensitivity operations, the response latency has been reduced from 642.7 ms to 96.5 ms, an improvement of 85.0%, and at the same time, the security verification accuracy has increased by 0.3%. More importantly, the standard deviation of the response latency has been significantly reduced, indicating that the response of the system is more stable and predictable.

[0192] During the market peak period (such as the 10 minutes before the opening and the period when major financial news is released), the peak number of voice commands processed by the system reaches 3,200 per second. Among them, the response delay of 92.7% of the commands is controlled within 100 ms, and the average delay for highly sensitive operations is 107.3 ms, verifying the performance of the system under extreme loads.

[0193] To verify the elasticity and fault tolerance of the system, different proportions of node failure scenarios were simulated in the actual deployment environment to test the system's recovery ability and performance retention rate:

[0194] The node failure data is shown in the following table:

[0195]

[0196] In the extreme case of 50% node failure, the function recovery time of the traditional redundant backup scheme is 95.3 seconds, and the system performance can only be maintained at 42.1%. While the function recovery time of this embodiment is only 12.5 seconds, and the system performance retention rate reaches 85.3%. This proves that the distributed elastic computing fault tolerance mechanism of this embodiment can effectively cope with large-scale node failures and significantly improve the stability and reliability of the system.

[0197] During the three-month actual production environment test, the system experienced 38 node failures and 12 network fluctuation events. All events were successfully processed without manual intervention, and the system service availability reached 99.997%, far higher than the industry standard level.

[0198] The above verification data fully proves that the distributed edge security symbiotic display voice interaction network of this embodiment has breakthroughly solved the inherent contradiction between response speed and security, and at the same time significantly improved the elasticity and fault tolerance of the system, providing a stable and reliable voice interaction solution for key business scenarios in the financial industry.

[0199] The system performance index set is shown in the following table:

[0200] System performance metric set P <![CDATA[p1: Computation latency]]> Normalized value range: 0.1 - 1.0 <![CDATA[p2: Throughput]]> Normalized value range: 0.2 - 0.9 <![CDATA[p3: Resource utilization rate]]> Normalized value range: 0.3 - 0.8 <![CDATA[p4: Parallel processing ability]]> Normalized value range: 0.2 - 0.95 <![CDATA[p5: Real-time responsiveness]]> Normalized value range: 0.15 - 0.9

[0201] The security index set data is shown in the following table:

[0202] Safety metric set S <![CDATA[s1: Voiceprint verification accuracy rate]]> Normalized value range: 0.6 - 0.99 <![CDATA[s2: Instruction anti-tampering ability]]> Normalized value range: 0.7 - 0.95 <![CDATA[s3: Data transmission encryption strength]]> Normalized value range: 0.8 - 0.99 <![CDATA[s4: Session Hijacking Protection Level]]> Normalized value range: 0.75 - 0.98 <![CDATA[s5: Abnormal behavior detection rate]]> Normalized value range: 0.65 - 0.9

[0203] The correlation weight matrix is shown in the following table:

[0204]

[0205] The security calculation entropy value H(P, S) = 0.68 calculated 10 minutes before the market opening (when the system load is low), and its theoretical maximum value H max= log(5 × 5) = 3.22. The complete data output by the secure computing entropy model includes the secure computing entropy value, the correlation weight matrix, and the entropy balance constraint parameters, providing a basis for subsequent resource allocation.

[0206] The data of the optimal resource allocation strategy is shown in the following table:

[0207]

[0208] In the application scenario, the utility function calculation unit calculates the utility value of each node in real time and reaches the Nash equilibrium point through 10 rounds of iteration (32 ms per round on average), generating a complete optimal resource allocation strategy matrix.

[0209] An example of the weight coefficient value data is shown in the following table:

[0210]

[0211] The system dynamically adjusts the weight coefficient according to the three-factor balance model and the characteristics of the trading period, increasing the weight of computing performance during the trading peak period and increasing the weight of security performance during the period with many sensitive operations.

[0212] An example of the processing strategy allocation data is shown in the following table:

[0213]

[0214] The system dynamically selects the processing strategy according to the instruction security sensitivity to achieve the optimal balance between security and response speed; for some parallel strategies, the result release table defines the result release method under different verification progress:

[0215] An example of the result release data is shown in the following table:

[0216]

[0217] An example of the health monitoring data is shown in the following table:

[0218]

[0219] When the system detects that Node_152 and Node_203 are in a critical state, the task migration mechanism is triggered. The following table shows some task migration plans:

[0220] An example of the task migration data is shown in the following table:

[0221] Example of task migration plan Source node Target node Degree of match Task_1207 Node_152 Node_078 0.86 Task_1209 Node_152 Node_187 0.92 Task_1211 Node_152 Node_042 0.79 Task_1356 Node_203 Node_187 0.85 Task_1358 Node_203 Node_042 0.88

[0222] An example of the adaptive security configuration data is shown in the following table:

[0223]

[0224] Through real-time risk assessment, the system automatically adjusts the security configuration when detecting changes in the risk level to ensure the optimization of resource usage in different security requirement scenarios.

[0225] The above describes the embodiments of the present invention. However, these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative rather than restrictive. Under the inspiration of these embodiments, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of these embodiments.

Claims

1. A method for voice interaction of a display, characterized in that, Including: Quantitatively characterizing data through the relationship between distributed game theory and computational security performance, and generating an optimal resource allocation strategy matrix; Generating a weight coefficient set through the secure computing entropy model data, the optimal resource allocation strategy matrix, and the three-factor balance model; Generating a dual-channel parallel execution strategy table and a result release strategy table through the weight coefficient set and the dual-channel parallel processing system; Generating a distributed elastic computing fault tolerance solution data packet through the dual-channel parallel execution strategy table and the result release strategy table; Generating an adaptive security configuration data set through the secure computing entropy model data, the dual-channel parallel execution strategy table, the result release strategy table, and the distributed elastic computing fault tolerance solution data packet.

2. The method for voice interaction of a display according to claim 1, characterized in that, In the steps of constructing the secure computing entropy model, the secure computing entropy model includes: an entropy calculation core unit, a correlation weight matrix unit, and an entropy balance constraint unit; the entropy balance constraint unit applies an entropy balance constraint algorithm to the secure computing entropy value to ensure that the secure computing entropy value does not exceed the theoretical threshold, and determines the theoretical threshold of the secure computing entropy value using the sizes of the performance metric set and the security metric set to ensure that the system operates within a reasonable range and maintains the balance between security and computing efficiency.

3. The method for voice interaction of a display according to claim 1, wherein In the steps of generating the optimal resource allocation strategy matrix by distributed game theory, the distributed non-zero-sum game model consists of an agent model unit, a utility function calculation unit, a Nash equilibrium solving unit, and a strategy optimization unit to form a utility function. The calculation formula for the utility function value of each agent in the agent model unit is as follows: Among them, U i represents the utility function value of the agent Player i ; a i represents the strategy of the agent Player i ; represents the set of strategies of other agents except the agent Player i ; α i is a coefficient related to the performance of the agent Player i ; is the performance function value of the agent Player i when taking the strategy a i and other agents take the strategy set a -i ; β i is a coefficient related to the security of the agent Player i ; is the security function value of the agent Player i when taking the strategy a i and other agents take the strategy set a1.

4. A method for voice interaction of a display according to claim 1, characterized in that, In the steps of generating the weight coefficient set by the three-factor balance model, the three-factor balance model consists of an efficiency evaluation unit, a cooperation factor calculation unit, and a weight optimization unit to form an efficiency evaluation unit. The calculation formula for the overall system efficiency value is as follows: E(C, S) = α(t)·P c (C) + β(t)·P s (S) + γ(t)·Syn(C, S); Among them, E(C, S) represents the overall system efficiency value; C is the computing resource allocation vector; S is the security resource allocation vector; α(t), β(t), and γ(t) are time-varying weight coefficients, where α(t) is used to measure the computing performance function P corresponding to the computing resource allocation vector C c (C)'s weight in the overall system efficiency calculation; β(t) is used to measure the security performance function P corresponding to the security resource allocation vector S s (S)'s weight in the overall system efficiency calculation; γ(t) is used to measure the weight of the synergy factor Syn(C, S) in the overall system efficiency calculation; P c (C) is the computing performance function related to the computing resource allocation vector C; P s (S) is the security performance function related to the security resource allocation vector S; Syn(C, S) is the synergy factor.

5. The method for voice interaction of a display according to claim 4, wherein The calculation formula for the cooperation factor is as follows: Among them, Syn(C, S) represents the cofactor value; C is the computing resource allocation vector, and C i is the i-th element in the computing resource allocation vector C, representing the allocation amount of the i-th computing resource; S is the security resource allocation vector, and S j is the j-th element in the security resource allocation vector S; n is the number of elements in the computing resource allocation vector C; m is the number of elements in the security resource allocation vector S; δ ij is the computing resource C i and the security resource S j is the cooperation coefficient between them; (1 - (C i - S j )) is used to adjust the impact of the difference in the allocation amounts of computing resources and security resources on the cofactor value.

6. The method for voice interaction of a display according to claim 1, characterized in that, In the steps of generating the dual-channel parallel execution strategy table and the result release strategy table by the dual-channel parallel processing system, the dual-channel parallel processing system consists of an instruction feature extraction unit, a security sensitivity calculation unit, a strategy generation unit, and a result release unit. The calculation formula for the security sensitivity value is as follows: Among them, S(V i ) represents the security sensitivity value of the instruction; V i represents the instruction feature vector; k is the number of elements in the instruction feature vector V i ; w j is the feature weight coefficient; v j is the j-th element in the instruction feature vector V i , representing the value of the j-th instruction feature; R j is the basic risk coefficient. By multiplying and accumulating these parameters, the security sensitivity value of the instruction is obtained.

7. The method for voice interaction of a display according to claim 6, characterized in that, The strategy generation unit applies a decision function Strategy(V i ) to generate dual-channel processing strategies, including a fully parallel strategy, a partially parallel strategy, and a verification-first strategy, and selects the corresponding strategy according to the comparison result between the instruction security sensitivity value and the preset threshold.

8. A method for voice interaction of a display according to claim 1, characterized in that, In the steps of generating the distributed elastic computing fault tolerance solution data packet, the distributed elastic computing fault tolerance system consists of a node health monitoring unit, a status classification unit, a task matching degree calculation unit, a task allocation unit, and a task migration unit. The calculation formula for the health index is as follows: H i = ω1·CPU i + ω2·MEM i + ω3·NET i + ω4·RESP i ; Among them, H i represents the health index of the i-th computing node; ω1, ω2, ω3, and ω4 are the weight coefficients corresponding to CPU i , MEM i , NET i , and RESP i respectively; CPU i represents the normalized CPU load of the node; MEM i represents the normalized memory usage rate of the node; NET i represents the normalized network connection quality of the node. The normalized network connection quality data is used to reflect the network status in the calculation of the health index; RESP i represents the normalized response time of the node.

9. A method for voice interaction of a display according to claim 8, characterized in that The task allocation unit applies an optimal task allocation algorithm to generate a task allocation plan. The calculation formula for its optimization objective function is as follows: Among them, A* represents the optimal task allocation scheme; A is the set of all possible task allocation schemes, and argmax A represents finding the task allocation scheme in set A that maximizes the value of the subsequent summation expression; m is the number of tasks, and n is the number of nodes; A ji represents the variable indicating whether task T j is allocated to node N i ; Match(T j , N i ) is the matching degree function between task T j and node N i , which is used to evaluate the adaptability between task T j and node N i . By multiplying A ji with Match(T j , N i ) and performing a double summation over all tasks and nodes, the optimal task allocation scheme is found.

10. A voice interaction system for a display, characterized in that, A display voice interaction method for executing any one of claims 1-9, including: A strategy matrix module that quantitatively characterizes data through the relationship between distributed game theory and computational security performance and generates an optimal resource allocation strategy matrix; A weight coefficient module that generates a weight coefficient set through the secure computing entropy model data, the optimal resource allocation strategy matrix, and the three-factor balance model; A processing strategy table module that generates a dual-channel parallel execution strategy table and a result release strategy table through the weight coefficient set and the dual-channel parallel processing system; A solution data module that generates a distributed elastic computing fault tolerance solution data packet through the dual-channel parallel execution strategy table and the result release strategy table; The security configuration module generates an adaptive security configuration data set by means of the security calculation entropy model data, the dual-channel parallel execution policy table, the result release policy table, and the distributed elastic computing fault tolerance solution data packet.

Images