Decision-making method for existing parameter hybrid drive

Abstract

The invention discloses a decision-making method for existing parameter hybrid driving, particularly relates to the field of simulated confrontation, and is used for solving the problem of group collaboration failure caused by dynamic conflicts of vision and text parameters, and solving the problem of group collaboration failure through adaptive downsampling and phase compensation when fusion of high-frequency vision parameters and low-frequency text parameters is processed. The feature counteracting effect in nonlinear fusion is reduced; meanwhile, on the basis of conjoint analysis of time-frequency energy distribution and causal association strength, a weight suppression matrix is constructed to weaken strategy contradictions, a strategy causal path is strengthened, and coordination consistency of decision instructions is guaranteed; besides, through cooperation of the hierarchical neural network and a time-space gating mechanism, dynamic fusion of an action sequence and strategy constraints is realized, and response speed and strategy consistency are improved. And finally, nonlinear oscillation of the instruction stream is inhibited in real time by means of damping control, group strategy deviation is prevented, and thus the overall decision quality is optimized.

Description

Technical Field

[0001] The present invention relates to the field of simulation confrontation, and more specifically, to a decision-making method for hybrid driving based on existing parameters. Background Art

[0002] The dynamic conflict of multi-modal decision-making parameters in the non-linear fusion process leads to the failure of group cooperation. In the confrontation simulation scenario, the defensive strategy parameters generated by the visual perception module based on the real-time environmental features and the aggressive strategy parameters parsed by the text semantic module from historical instructions, when fused through a conventional gating network, due to the essential differences in the dynamic characteristics of the parameters, an unexpected feature cancellation effect occurs. Its technical mechanism is manifested as follows: visual parameters are driven by local environmental changes and have the characteristic of high-frequency fine-tuning, while text parameters are constrained by global strategies and show low-frequency steady-state characteristics. The two form a phase misalignment during the time-series evolution process; when the gating network uses a fixed non-linear function (such as sigmoid) for parameter fusion, the gradient update directions of high-frequency parameters and low-frequency parameters generate adversarial interference within a specific time window, resulting in the fused decision vector continuously oscillating in the strategy space. This problem is exponentially amplified in the multi-agent cooperation scenario: the parameter oscillation of a single agent propagates through the strategy coupling graph in the group, forming a positive feedback loop, causing the group strategy to deviate from the equilibrium state; the decision-making system needs to repeatedly adjust the parameters to correct the deviation, resulting in the exhaustion of computing resources and a sharp increase in response delay, and finally causing the large-scale cooperative decision-making system to fall into a paralyzed state.

[0003] To solve the above problems, a technical solution is provided now. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a decision-making method for hybrid driving based on existing parameters. Through adaptive downsampling and phase compensation, the feature cancellation effect in non-linear fusion is reduced; at the same time, based on the joint analysis of time-frequency energy distribution and causal correlation strength, a suppression weight matrix is constructed to weaken the strategy contradiction and strengthen the strategy causal path, ensuring the coordination and consistency of decision-making instructions; in addition, through the cooperation of a hierarchical neural network and a spatio-temporal gating mechanism, the dynamic fusion of action sequences and strategy constraints is realized, improving the response speed and strategy consistency; finally, the non-linear oscillation of the instruction flow is suppressed in real time by means of damping control to prevent the deviation of the group strategy, thereby optimizing the overall decision-making quality to solve the problems raised in the above background art.

[0005] To achieve the above object, the present invention provides the following technical solutions: S1: Based on the mutation detection of the fluctuation characteristics of the visual parameter flow and the identification of the strategy instruction events of the text parameter flow, align the key conflict intervals of the multi-modal parameters through the dynamic time warping algorithm, and generate a set of non-uniformly distributed conflict period coordinates; S2: Implement event-driven adaptive downsampling on high-frequency visual parameters within the defined conflict period coordinates, align their time bases with the text parameters, and use time-frequency analysis technology to extract phase compensation vectors to eliminate the inter-modal phase difference; S3: Conduct a joint analysis of the time-frequency energy distribution and causal correlation strength of the synchronized parameters, construct a counterfactual intervention model to generate a frequency band suppression weight matrix, and directionally weaken the energy of the frequency bands that cause policy contradictions and strengthen the policy causal path; S4: Input the conflict-free visual parameters into the tactical layer pulse neural network to generate a millisecond-level action sequence, and input the text parameters into the policy layer memory-enhanced network to generate a constraint framework, and achieve multi-granularity instruction fusion through a spatio-temporal gating mechanism; S5: Real-time monitor the non-linear oscillation characteristics of the fused instruction stream, construct a damping controller based on the Lyapunov stability theory, and inject a negative feedback correction signal to maintain the convergence of the group strategy when detecting the characteristics of the edge of chaos.

[0006] In a preferred embodiment, step S1 includes the following contents: First, detect the mutation points of the fluctuation characteristics by calculating the local fluctuation entropy of the visual parameter stream, statistically calculate the probability distribution of the change amplitude of the visual parameter stream within each time window and calculate the information entropy, and mark the central time point of the corresponding time window as the visual mutation point when the local fluctuation entropy exceeds the preset entropy threshold; Then, use the exponential smoothing cumulative deviation method to identify the policy instruction events of the text parameter stream, apply exponential smoothing processing to the text parameter stream to reduce short-term fluctuations, calculate the dynamic deviation between the smoothed parameter and the moving average value, and mark the corresponding time point as the policy instruction event point when the dynamic deviation exceeds the preset deviation threshold; Next, use the dynamic time warping algorithm to perform temporal alignment on the set of visual mutation points and the set of policy instruction event points, construct a cumulative distance matrix and solve the minimum cumulative distance path through dynamic programming to generate a set of aligned time point pairs; Finally, generate a set of conflict period coordinates based on the set of aligned time point pairs, define the conflict period for each pair of aligned time points and perform expansion and merging to form a non-uniformly distributed set of conflict period coordinates.

[0007] In a preferred embodiment, step S2 includes the following contents: Within each conflict period, first calculate the local change rate of the high-frequency visual parameter stream to identify the event trigger points, and mark the corresponding time points as event trigger points by judging that the local change rate exceeds the preset change rate threshold; Then, perform adaptive downsampling based on the event trigger points, retain the parameter values at the event trigger points, and use linear interpolation for non-event trigger points to generate approximate values, so that the number of sampling points of the downsampled high-frequency visual parameter stream within the conflict period is the same as that of the low-frequency text parameter stream within the same conflict period, achieving time base alignment.

[0008] In a preferred embodiment, step S2 further includes the following: Next, perform continuous wavelet transform on the downsampled high-frequency visual parameter stream and low-frequency text parameter stream respectively to generate their time-frequency spectra, and calculate the phase difference between the two at each time point and scale; construct a phase compensation vector based on the phase difference, and take the negative value of the phase difference at the dominant scale of each time point as the compensation phase; Finally, apply phase rotation adjustment to the downsampled high-frequency visual parameter stream, multiply the complex signal at each time point by the complex exponential term, where the phase angle of the complex exponential term is the compensation phase value, and take the real part to generate the synchronized parameter stream, thus achieving alignment in time and phase with the low-frequency text parameter stream.

[0009] In a preferred embodiment, step S3 includes the following: Perform wavelet packet decomposition on the synchronized visual parameter stream to generate a time-frequency energy matrix, and perform short-time Fourier transform on the text parameter stream to generate a main frequency energy feature matrix; Calculate the cross-coherence matrix based on the time-frequency energy matrix of the visual parameter stream and the time-frequency energy matrix of the text parameter stream to quantify the energy coupling strength of the physical layer; Construct a structural equation model to analyze the causal effect of the visual parameter stream on the text parameter stream, and calculate the conditional entropy change value through counterfactual reasoning to characterize the causal association strength of the logic layer; Fuse the cross-coherence matrix, the sign of the causal strength coefficient, and the conditional entropy change value to generate a dynamic suppression weight matrix, apply suppression to the frequency bands with high coherence and negative causal association through an exponential decay function, and apply positive compensation to the policy main frequency at the same time; Finally, perform time-frequency domain filtering on the synchronized visual parameter stream to generate a conflict-free visual parameter stream, so as to weaken the energy of the frequency bands that cause policy contradictions and strengthen the policy causal path.

[0010] In a preferred embodiment, step S4 includes the following: Perform dynamic threshold modulation on the deconflicted visual parameter stream through a tactical layer pulsed neural network, and use a non-linear neuron model to simulate the tactical action generation process. Among them, the evolution of the neuron membrane potential over time is driven by the product of the deconflicted visual parameter stream and the phase modulation term. At the same time, the dynamic threshold is adjusted according to the local energy accumulation of the deconflicted visual parameter stream. When the neuron membrane potential exceeds the dynamic threshold, a pulse signal is triggered and the membrane potential is reset. The pulse sequence is mapped to a millisecond-level action sequence through a multi-layer structure.

[0011] In a preferred embodiment, step S4 includes the following content: Perform attention-enhanced bidirectional cyclic processing on the text parameter stream through a policy layer memory-enhanced network. The update of the hidden state is jointly determined by the current text parameter stream input, the previous moment's hidden state, and the weighted summary of the historical hidden states. Among them, the weighted summary of the historical hidden states is calculated through an attention attenuation factor, and finally a constraint framework is generated through a sigmoid activation function.

[0012] In a preferred embodiment, step S4 includes the following content: Realize the dynamic fusion of the millisecond-level action sequence and the constraint framework through a spatio-temporal gating mechanism, construct a spatio-temporal gating unit to calculate the dynamic fusion coefficient. Among them, the dynamic fusion coefficient is determined by the millisecond-level action sequence, the constraint framework, and their spatio-temporal correlation term. The spatio-temporal correlation term is calculated through the product integration of the millisecond-level action sequence and the constraint framework within a local time window. Finally, the fusion instruction stream is generated by the weighted average of the millisecond-level action sequence and the constraint framework.

[0013] In a preferred embodiment, step S5 includes the following content: Perform dynamic characteristic analysis on the fusion instruction stream through phase space reconstruction technology, select the embedding dimension and time delay parameters to construct a reconstruction vector, and calculate the largest Lyapunov exponent in the reconstructed phase space to judge the chaotic behavior of the fusion instruction stream; Design a damping controller through Lyapunov stability theory, define a Lyapunov function and design the dynamic evolution of the damping term to ensure the stability of the fusion instruction stream; Monitor the largest Lyapunov exponent through a chaos edge detection mechanism. When its absolute value is less than the chaos edge threshold, it is determined that the fusion instruction stream is at the chaos edge, and a damping term is injected as a negative feedback correction signal to generate a corrected instruction stream to maintain the convergence of the group strategy.

[0014] The technical effects and advantages of the present invention facing the existing decision-making method driven by parameter mixing: Through the dynamic conflict period intelligent focusing mechanism, the present invention accurately locks the core time window where high-frequency strategy contradictions occur in complex confrontation scenarios, compressing the computational resource consumption during non-essential periods; adopts cross-modal spatio-temporal reference synchronization technology to eliminate the timing misalignment and phase deviation between visual and text parameters, establishing a precise alignment basis for multi-source heterogeneous data; through a two-dimensional analysis mechanism of time-frequency physical characteristics and causal logic correlation, breakthroughly identifies cross-modal invisible conflict frequency bands that are difficult to detect by traditional methods, and systematically solves the deep mutual exclusion problem of defensive and aggressive strategy parameters at the levels of energy coupling and causal conduction; based on a hierarchical decision-making architecture of pulsed neural networks and memory-enhanced networks, realizes the dynamic balance between millisecond-level tactical micro-operation response and minute-level strategy framework stability, ensuring that high-frequency actions are executed without deviating from macroscopic constraint conditions; relies on the non-linear stability theory to construct a group decision-making oscillation propagation blocking mechanism, and through real-time analysis of strategy waveforms and closed-loop negative feedback control, effectively inhibits the chain diffusion of local parameter conflicts in the collaborative network, forming an all-round guarantee system with precise and controllable tactical execution, dynamically corrected strategy goals, and steady convergence of group states, providing intelligent decision-making support with both agile response speed and continuous robustness for dynamic confrontation environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic flowchart of the decision-making method of the present invention for the existing parameter hybrid drive. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0017] Embodiment 1: Figure 1 The decision-making method of the present invention for the existing parameter hybrid drive is given, including: S1: Detect the mutation of the fluctuation characteristics of the visual parameter stream and identify the strategy instruction events of the text parameter stream, align the key conflict intervals of multi-modal parameters through the dynamic time warping algorithm, and generate a set of non-uniformly distributed conflict period coordinates.

[0018] S2: Within the defined conflict period coordinates, perform event-driven adaptive downsampling on high-frequency visual parameters to align their time reference with text parameters, and use time-frequency analysis technology to extract phase compensation vectors to eliminate the phase difference between modalities.

[0019] S3: Conduct a joint analysis of the time-frequency energy distribution and causal association strength of the synchronized parameters, construct a counterfactual intervention model to generate a frequency band suppression weight matrix, and directionally weaken the energy of the frequency band that causes policy contradictions and strengthen the policy causal path.

[0020] S4: Input the conflict-resolved visual parameters into the tactical layer pulsed neural network to generate a millisecond-level action sequence, and input the text parameters into the policy layer memory-enhanced network to generate a constraint framework, and achieve multi-granularity instruction fusion through a spatio-temporal gating mechanism.

[0021] S5: Real-time monitor the non-linear oscillation characteristics of the fused instruction stream, construct a damping controller based on the Lyapunov stability theory, and inject a negative feedback correction signal to maintain the convergence of the group strategy when detecting the characteristics of the edge of chaos.

[0022] In a multi-modal decision-making scenario, due to the differences in their respective dynamic characteristics, the visual parameter stream and the text parameter stream are prone to conflicts in the time series evolution, affecting the efficiency of group collaborative decision-making. Step S1 focuses on detecting the mutation of the fluctuation characteristics of the visual parameter stream and identifying the policy instruction events of the text parameter stream, aligning the key conflict intervals of the multi-modal parameters through the dynamic time warping algorithm, and generating a set of conflict period coordinates with non-uniform distribution. This set provides an accurate time window for the subsequent steps to achieve parameter synchronization and conflict resolution.

[0023] Step S1 includes the following content: S1.1, Detection of the mutation of the fluctuation characteristics of the visual parameter stream: Detect the mutation of the fluctuation characteristics of the visual parameter stream. The visual parameter stream refers to the defensive strategy parameters generated by the visual perception module in the time series. First, calculate the change amplitude between adjacent time points of the visual parameter stream. Specifically, for each time point, calculate the absolute value of the difference between the parameter value at the current time point and the parameter value at the previous time point to reflect the dynamic change trend of the visual parameter stream. Next, introduce a sliding window mechanism to analyze the local characteristics within each time window. The sliding window covers a certain number of consecutive time points, and the probability distribution of all change amplitudes is statistically analyzed within the window. The specific method is to divide the change amplitudes within the window into several intervals, calculate the frequency of each interval, and then calculate the information entropy based on these frequencies, which is called the local fluctuation entropy. The local fluctuation entropy reflects the complexity and uncertainty of the fluctuation of the visual parameter stream within the window. If the local fluctuation entropy of a certain time window exceeds the pre-set entropy threshold, the central time point of this window is marked as a mutation point, indicating that a significant change has occurred in the fluctuation characteristics of the visual parameter stream. Finally, collect all the marked mutation points to form a visual mutation point set as the input for subsequent processing.

[0024] Using local fluctuation entropy to detect mutation points can fully capture the dynamic characteristics of the visual parameter flow within a local range. Compared with the method based only on the change amplitude threshold, local fluctuation entropy comprehensively considers the uncertainty and complexity of fluctuations and can more accurately identify significant changes in the fluctuation pattern, thereby improving the accuracy of mutation point detection.

[0025] S1.2, Strategy instruction event recognition of the text parameter flow: Perform strategy instruction event recognition on the text parameter flow. The text parameter flow refers to the aggressive strategy parameters generated by the text semantic module parsing historical instructions in the time series. First, calculate the dynamic deviation of the text parameter flow. The specific method is to first apply exponential smoothing to the text parameter flow to generate a smoothed parameter sequence to reduce the interference of short-term fluctuations. Then, calculate the moving average of the smoothed parameter sequence to reflect the long-term trend of the parameters. For each time point, calculate the absolute difference between the smoothed parameter value and the moving average, which is called the dynamic deviation. The magnitude of the dynamic deviation reflects the degree to which the text parameter flow deviates from the trend. Next, set a deviation threshold. When the dynamic deviation at a certain time point exceeds the deviation threshold, mark this time point as a strategy instruction event point, indicating that a significant strategy turn has occurred in the text parameter flow at this point. Finally, collect all the marked event points to form a text event point set.

[0026] Using exponential smoothing cumulative deviation to identify strategy instruction events can effectively smooth the short-term fluctuations in the text parameter flow and highlight the changes in the long-term trend. This method is particularly suitable for detecting instruction events at the strategy level, avoiding misjudging minor adjustments at the tactical level as key events, thereby improving the pertinence and accuracy of event recognition.

[0027] S1.3, Aligning conflict intervals using the dynamic time warping algorithm: Use the dynamic time warping algorithm to perform temporal alignment on the visual mutation point set and the text event point set. The visual mutation point set and the text event point set are respectively from the outputs of the previous two steps, representing the key time points of the two parameter flows. First, regard the visual mutation point set and the text event point set as two independent time series. Then, define the distance between time points as the absolute time difference between the two, and construct a cumulative distance matrix based on this. Each element of the cumulative distance matrix represents the sum of the time differences in all possible paths from the starting point to the current point. Subsequently, solve the minimum cumulative distance path in the cumulative distance matrix through the dynamic programming method. Specifically, from the starting point to the end point of the matrix, gradually select the path with the minimum cumulative distance to ensure the lowest overall alignment cost. After completing the path solution, backtrack this path to generate the correspondence between visual mutation points and text event points, forming an aligned time point pair set as the input for subsequent processing.

[0028] The dynamic time warping algorithm can handle the non - linear temporal misalignment between the visual parameter stream and the text parameter stream, and adapt to the dynamic characteristic differences between the two. By calculating the minimum cumulative distance path, it ensures the optimal temporal matching of the alignment result, thereby accurately identifying the conflict intervals.

[0029] S1.4, generate the conflict period coordinate set: Generate the conflict period coordinate set according to the alignment time point pair set. The alignment time point pair set contains the corresponding relationship between visual mutation points and text event points. First, for each pair of alignment time points, define the conflict period as the intersection of a certain time range around the visual mutation point and the text event point. Specifically, with each time point as the center, expand the preset time width forward and backward to obtain their respective period ranges, and then take the overlapping part as the initial conflict period. Then, check the conflict periods of all alignment time point pairs. If the time difference between a certain pair of time points is less than the preset range, merge the corresponding conflict periods into an extended period to cover the continuous conflict area. Finally, further merge all overlapping conflict periods to form a non - uniformly distributed conflict period coordinate set, which is the final output of step S1.

[0030] Through the period expansion and merging operations, ensure that the conflict period coordinate set covers all potential conflict points, while avoiding redundancy caused by overly scattered periods. The non - uniformly distributed time intervals generated by this method can provide an accurate and efficient time window for subsequent parameter synchronization and conflict resolution, improving the pertinence of processing.

[0031] In step S1, mutation points are detected by calculating the local fluctuation entropy of the visual parameter stream, and the strategic instruction events of the text parameter stream are identified by combining the exponentially smoothed cumulative deviation. The dynamic time warping algorithm is used to align the visual mutation points and text event points, and finally a non - uniformly distributed conflict period coordinate set is generated. This set provides a reliable time benchmark for subsequent processing, ensuring the accuracy and efficiency of multi - modal parameter conflict resolution.

[0032] In step S1, through the mutation detection of the fluctuation characteristics of the visual parameter stream and the identification of the strategic instruction events of the text parameter stream, combined with the dynamic time warping algorithm, a non - uniformly distributed conflict period coordinate set is generated. This set accurately marks the key time intervals where dynamic conflicts occur in the temporal sequence between the visual parameter stream and the text parameter stream. Based on this set, step S2 addresses the temporal misalignment and phase deviation problems between the high - frequency visual parameter stream and the low - frequency text parameter stream within the conflict period. Through event - driven adaptive downsampling and phase compensation techniques, it realizes the alignment of their time benchmarks and phase synchronization, laying a data foundation for the time - frequency energy distribution and causal correlation analysis in step S3.

[0033] Step S2 includes the following: S2.1, Event-driven Adaptive Downsampling: Perform event-driven adaptive downsampling on the high-frequency visual parameter stream during the conflict period to adjust its time base to match the sampling density of the low-frequency text parameter stream. The conflict period is determined by the previous step S1, which identifies the key interval where dynamic conflicts occur between the high-frequency visual parameter stream and the low-frequency text parameter stream in terms of time. Since the sampling rate of the high-frequency visual parameter stream is much higher than that of the low-frequency text parameter stream, there is a misalignment in the corresponding relationship between the two at time points. Therefore, first calculate the local change rate of the high-frequency visual parameter stream at each time point. The specific method is as follows: take the difference between the parameter values of two adjacent time points and divide it by the time interval between these two time points to obtain the change speed of the parameter value over time. Then, set a predefined change rate threshold. When the local change rate at a certain time point exceeds this threshold, mark this time point as an event trigger point, indicating that a significant change has occurred in the high-frequency visual parameter stream at this point. Then, perform downsampling based on these event trigger points: retain the parameter values at all event trigger points. For the time points between two adjacent event trigger points, generate approximate values using linear interpolation. The calculation method of linear interpolation is to calculate the parameter value of the intermediate time point according to the time positions and parameter values of the two event trigger points, in proportion to the time. Specifically, take the difference between the parameter values of the two event trigger points and allocate the interpolation result according to the proportion of the time distance between the intermediate time point and the trigger point. Finally, by adjusting the number of interpolation points, make the total number of sampling points of the downsampled high-frequency visual parameter stream in each conflict period consistent with the total number of sampling points of the low-frequency text parameter stream in the same conflict period, thus achieving alignment of their time bases.

[0034] Using event-driven adaptive downsampling can ensure that the key change information in the high-frequency visual parameter stream is retained, avoid the loss of important features due to downsampling, and at the same time reduce the sampling points in non-critical areas through linear interpolation, thereby reducing the computational complexity of subsequent processing. Compared with traditional uniform downsampling, this method can better adapt to the dynamic characteristics of the parameter stream, and the generated downsampling result is accurately matched with the low-frequency text parameter stream in terms of time distribution.

[0035] S2.2, Time-frequency Analysis and Phase Compensation Vector Extraction: Use time-frequency analysis technology to extract the phase compensation vector to eliminate the phase difference between the high-frequency visual parameter stream and the low-frequency text parameter stream after downsampling, and further improve the synchronization accuracy. During each conflict period, first perform continuous wavelet transform on the downsampled high-frequency visual parameter stream and low-frequency text parameter stream respectively to generate their respective time-frequency spectra. The calculation process of continuous wavelet transform is as follows: convolve the parameter stream with a set of predefined wavelet basis functions, and by adjusting the scale parameter and time translation parameter of the wavelet basis functions, decompose the parameter stream into coefficients at different time positions and different frequency scales. These coefficients reflect the energy distribution of the parameter stream in the time-frequency domain. Then, based on the generated time-frequency spectra, calculate the phase difference between the downsampled high-frequency visual parameter stream and the low-frequency text parameter stream at the same time position and the same frequency scale. The specific method is as follows: take the phase angles of the coefficients of the two at each time point and scale, and calculate the difference between the two phase angles. Then, construct the phase compensation vector: for each time point of the downsampled high-frequency visual parameter stream, determine its dominant scale, that is, the scale with the largest energy among all scales, and the energy size is calculated by the square of the amplitude of the coefficient; at this dominant scale, take the negative value of the phase difference as the compensation phase value at this time point; finally, combine the compensation phase values of all time points in chronological order to form the phase compensation vector.

[0036] Continuous wavelet transform can simultaneously analyze the characteristics of the parameter stream in time and frequency. Compared with the traditional method that only analyzes frequency, it is more suitable for dealing with the dynamic changes of non-stationary signals. By calculating the phase difference at the dominant scale and generating the phase compensation vector, the phase deviation of the downsampled high-frequency visual parameter stream and the low-frequency text parameter stream at the key frequencies can be corrected specifically. This method improves the accuracy of phase analysis, ensures the synchronization consistency of the two in the time-frequency domain, and provides a high-quality phase correction basis for generating the synchronized parameter stream.

[0037] S2.3, Phase adjustment and synchronized parameter generation: Based on the extracted phase compensation vector, perform phase adjustment on the downsampled high-frequency visual parameter stream to generate the final synchronized parameter stream, achieving complete alignment with the low-frequency text parameter stream in terms of time and phase. First, regard the downsampled high-frequency visual parameter stream as a complex signal, where the real part is the original parameter value and the initial value of the imaginary part is zero. Then, apply phase rotation adjustment to the complex signal at each time point. Specifically, multiply the complex signal at this time point by a complex exponential term, where the phase angle of the complex exponential term is taken from the compensation phase value at the corresponding time point of the phase compensation vector. This operation is equivalent to rotating the signal by an angle in the complex plane, and the size of the angle is determined by the compensation phase value. After rotation, extract the real part of the complex signal as the parameter value after phase adjustment. Finally, the high-frequency visual parameter stream after phase adjustment is aligned with the low-frequency text parameter stream in terms of time reference (guaranteed by step S2.1) and is also consistent with the low-frequency text parameter stream in terms of phase, forming a synchronized parameter stream.

[0038] Through phase rotation adjustment, it is possible to accurately correct the phase deviation of the downsampled high-frequency visual parameter stream without changing its amplitude, ensuring phase synchronization with the low-frequency text parameter stream. It avoids the amplitude distortion that may be introduced by directly modifying the parameter value, retains the original dynamic characteristics of the parameter stream, and at the same time achieves high-precision alignment of the cross-modal parameter streams in terms of time and phase.

[0039] Step S2 realizes the synchronization processing of the high-frequency visual parameter stream and the low-frequency text parameter stream through three sub-steps: First, use event-driven adaptive downsampling to adjust the time reference of the high-frequency visual parameter stream so that the number of sampling points matches that of the low-frequency text parameter stream; then extract the time-frequency spectrum and calculate the phase compensation vector through continuous wavelet transform to eliminate the phase difference between the two; finally, perform phase rotation based on the phase compensation vector to generate a synchronized parameter stream that is aligned with the low-frequency text parameter stream in both time and phase. This process ensures the accuracy and integrity of the synchronized data and provides high-quality input for subsequent processing.

[0040] In step S2, through event-driven adaptive downsampling and alignment of the time reference of the high-frequency visual parameter stream, as well as phase compensation technology, the phase difference between the visual parameter stream and the low-frequency text parameter stream during the conflict period is eliminated, generating the synchronized visual parameter stream and text parameter stream. The set of conflict period coordinates is provided by step S1, which identifies the key conflict intervals of the multi-modal parameters. Based on this, step S3 performs a joint analysis of the time-frequency energy distribution and causal association strength for the synchronized visual parameter stream and text parameter stream, constructs a counterfactual intervention model to generate a frequency band suppression weight matrix, directionally weakens the energy of the frequency band that causes policy contradictions and strengthens the policy causal path, and provides a conflict-free visual parameter stream for step S4.

[0041] Step S3 includes the following content: S3.1, Time-frequency energy distribution extraction: Perform time-frequency energy distribution extraction on the synchronized visual parameter stream and text parameter stream to capture their dynamic characteristics in time and frequency. For the synchronized visual parameter stream, first perform wavelet packet decomposition. Wavelet packet decomposition generates a time-frequency energy matrix by decomposing the signal into different levels, which records the energy distribution of the visual parameter stream at different time points and different frequency levels. The energy value is calculated by the square of the coefficients of each level after wavelet packet decomposition. The Daubechies wavelet basis function is used for decomposition, and the decomposition level is determined according to a preset value to balance the requirements of time resolution and frequency resolution while controlling the computational complexity. For the synchronized text parameter stream, perform short-time Fourier transform to generate its main frequency energy feature matrix. Short-time Fourier transform calculates the energy distribution of the signal at different time periods and frequencies by sliding a Gaussian window on the time axis and performing Fourier transform on the signal within the window. The window length is adaptively adjusted according to the duration of the conflict period to ensure sufficient frequency resolution within the conflict period.

[0042] Wavelet packet decomposition is suitable for processing non-stationary signals and can provide high-precision time and frequency localization features, especially suitable for capturing high-frequency fluctuation characteristics in the visual parameter stream. Short-time Fourier transform provides the main frequency energy features of the text parameter stream while maintaining computational efficiency, facilitating comparison with the visual parameter stream. The combination of these two time-frequency analysis techniques comprehensively characterizes the dynamic behavior of the synchronized visual parameter stream and text parameter stream during the conflict period.

[0043] S3.2, Cross-coherence calculation: Based on the time-frequency energy matrix of the visual parameter stream and the main frequency energy feature matrix of the text parameter stream, calculate the cross-coherence between them at each time-frequency unit to quantify the energy coupling strength at the physical layer. The calculation process of cross-coherence is as follows: for each time-frequency unit, calculate the product of the energy value of the visual parameter stream and the energy value of the text parameter stream, and then divide it by the square root of the sum of the squares of the energy of the visual parameter stream and the energy of the text parameter stream, obtaining a value between 0 and 1. This value reflects the energy similarity between the visual parameter stream and the text parameter stream at this time-frequency unit. The larger the value, the more similar the energy distributions of the two at this time-frequency unit, and the higher the energy coupling strength. Finally, generate a cross-coherence matrix that records the coherence values of all time-frequency units.

[0044] As a dimensionless similarity measure, cross-coherence can effectively capture the energy correlation between the visual parameter stream and the text parameter stream in the time-frequency domain. Compared with traditional correlation measures, cross-coherence pays more attention to the similarity of energy distributions and is suitable for analyzing the physical layer coupling characteristics of multimodal signals.

[0045] S3.3, Causal association strength analysis: For the synchronized visual parameter stream and text parameter stream, construct a structural equation model to analyze the causal relationship between them, and quantify the causal association strength through counterfactual reasoning. First, assume that there is a linear causal effect of the visual parameter stream on the text parameter stream, and construct a linear regression model, where the text parameter stream is used as the dependent variable and the visual parameter stream is used as the independent variable, and estimate the regression coefficients by the least squares method. The regression coefficients characterize the direction and magnitude of the causal effect. Then, adopt the counterfactual reasoning method to simulate the intervention effect on the visual parameter stream. The specific process is as follows: Set the baseline value and intervention value of the visual parameter stream, and calculate the change in conditional entropy of the text parameter stream before and after the intervention. The conditional entropy reflects the change in the uncertainty of the text parameter stream given the intervention of the visual parameter stream. The larger the entropy change value, the higher the causal association strength. Finally, generate the causal strength coefficient and the conditional entropy change matrix, which provides a quantitative index for the causal association at the logical layer for generating the band suppression weights.

[0046] Combining the structural equation model with counterfactual reasoning can reveal the causal intervention effect of the visual parameter stream on the text parameter stream from the logical level. By calculating the change in conditional entropy, not only the strength of the causal relationship is quantified, but also a basis is provided for identifying the causal paths that trigger policy contradictions. This method helps to directionally weaken the bands with negative causal associations and improve the pertinence of conflict resolution.

[0047] S3.4, Generation of the band suppression weight matrix: Fuse the cross-coherence, causal strength coefficient, and conditional entropy change value to generate a dynamic suppression weight matrix to achieve directional weakening of the energy of the bands that trigger policy contradictions and energy compensation for the main policy frequency. The specific process is as follows: Construct an exponential decay function, with the input being the cross-coherence, the sign of the causal strength coefficient, and the conditional entropy change value, and the output being the suppression weight. For each time-frequency unit, calculate the product of the cross-coherence and the conditional entropy change value, then multiply it by the sign of the causal strength coefficient (1 for positive causality and -1 for negative causality), and take its negative value as the exponential term of the exponential decay function to calculate the suppression weight. The smaller the weight value, the stronger the suppression of the time-frequency unit. Especially for the bands with high cross-coherence and negative causal associations, the weight approaches 0, achieving significant energy decay. Then, for the predefined main policy frequency, apply a positive compensation on the basis of the suppression weight, specifically by increasing the weight value of the main policy frequency by a fixed compensation coefficient to strengthen the policy causal path. Finally, generate the band suppression weight matrix, which records the weight values of all time-frequency units.

[0048] The strategic main frequency refers to the frequency component in the text parameter stream that is directly related to the global strategic goal and carries the main strategic intention. The text parameter stream is generated by the text semantic module, reflecting the low-frequency steady-state characteristics of the aggressive strategy. The strategic main frequency is the key frequency band that has a dominant influence on group collaborative decision-making, usually manifested as a frequency band with concentrated and stable energy. The strategic main frequency is identified through the main frequency energy feature matrix extracted by the short-time Fourier transform, representing the core dynamic characteristics consistent with the strategic goal in the text parameter stream. Strengthening the energy of the strategic main frequency aims to ensure that the visual parameter stream after conflict resolution maintains coordination with the global strategic direction while eliminating strategic contradictions, thereby avoiding weakening the strategic orientation of the overall decision-making due to excessive inhibition.

[0049] By fusing the energy coupling index (cross coherence) of the physical layer and the causal association index (causal strength coefficient and conditional entropy change value) of the logical layer, a dynamic suppression weight matrix is generated, which can accurately identify and weaken the frequency bands that cause strategic contradictions while protecting and strengthening the frequency bands related to the strategy. This two-dimensional constraint mechanism ensures the pertinence and effectiveness of conflict resolution, avoiding the risks of blind suppression or overcompensation.

[0050] S3.5, Parameter conflict resolution processing: Based on the frequency band suppression weight matrix, perform time-frequency domain filtering on the synchronized visual parameter stream to generate a conflict-free visual parameter stream. The specific process is as follows: First, perform the Fourier transform on the synchronized visual parameter stream to obtain its frequency domain representation; then, multiply the frequency domain signal by the corresponding elements of the frequency band suppression weight matrix to adjust the energy of specific frequency bands; finally, perform the inverse Fourier transform on the adjusted frequency domain signal to convert it back to the time domain to obtain a conflict-free visual parameter stream.

[0051] Time-frequency domain filtering can orientedly adjust the energy of specific frequency bands without changing the overall structure of the visual parameter stream, achieving precise suppression of the frequency bands that cause strategic contradictions. Compared with traditional time-domain filtering, time-frequency domain filtering is more suitable for processing non-stationary signals, can flexibly adapt to the dynamic characteristics of the synchronized visual parameter stream, and ensure that the conflict-free visual parameter stream does not introduce strategic contradictions in the subsequent generation of tactical layer action sequences.

[0052] In step S3, by jointly analyzing the time-frequency energy distribution and causal association strength of the synchronized visual parameter stream and text parameter stream, a counterfactual intervention model is constructed, a frequency band suppression weight matrix is generated, and conflict resolution processing is performed on the visual parameter stream to obtain a conflict-free visual parameter stream. At the same time, the text parameter stream, as the input at the strategy level, is directly passed to step S4 without being processed. Based on this, step S4 uses the conflict-free visual parameter stream to generate a millisecond-level action sequence at the tactical level, uses the text parameter stream to generate a constraint framework at the strategy level, and realizes multi-granularity instruction fusion between the two through an innovative spatio-temporal gating mechanism, and finally outputs a fused instruction stream for use in step S5.

[0053] Step S4 includes the following: S4.1, the tactical layer spiking neural network generates a millisecond-level action sequence: For the conflict-free visual parameter stream, a millisecond-level action sequence is generated through the tactical layer spiking neural network SNN. The tactical layer spiking neural network adopts a dynamic threshold modulation mechanism and combines a non-linear neuron model to simulate the tactical action generation process. The membrane potential of the neuron evolves over time, and its rate of change is proportional to the product of the conflict-free visual parameter stream and the phase modulation term, and is also affected by the negative feedback of the membrane potential itself. The phase modulation term is derived from and normalized by the phase compensation vector in the previous step S2, and reflects the temporal characteristics of the conflict-free visual parameter stream. The threshold is dynamically adjusted based on the energy accumulation of the conflict-free visual parameter stream within a local time window, specifically as the reference threshold plus the weighted value of the integral of the absolute value of the conflict-free visual parameter stream within the local time window. When the membrane potential exceeds the dynamic threshold, the neuron emits a pulse signal and resets the membrane potential to a preset value. The tactical layer spiking neural network consists of a multi-layer structure, including an input layer, two hidden layers, and an output layer, and maps the pulse sequence to a millisecond-level action sequence through weighted summation and time delay mechanisms. The generation process of the action sequence is: weighted summation of the pulse signals of each neuron, and the weight and time delay parameters are determined according to network training, and finally a millisecond-level action sequence is output as the high-frequency action instruction at the tactical level.

[0054] The tactical layer spiking neural network simulates the efficient information processing method of the biological nervous system and is particularly suitable for processing high-frequency and time-sensitive tactical action generation tasks. The dynamic threshold modulation mechanism enables the network to adaptively adjust its sensitivity to the conflict-free visual parameter stream and enhances its response ability to key features therein. Combined with the phase modulation term, the network better captures the temporal dynamic characteristics of the conflict-free visual parameter stream, improves the generation accuracy and response speed of the millisecond-level action sequence. This processing method ensures the efficiency and real-time nature of tactical actions and provides an accurate input basis for subsequent multi-granularity instruction fusion.

[0055] S4.2, the strategy layer memory-enhanced network generates a constraint framework: For the text parameter stream, a constraint framework is generated through a policy layer memory-augmented network. The policy layer memory-augmented network adopts a bidirectional recurrent structure based on the attention mechanism to extract long-term dependence features in the text parameter stream. The update process of the hidden state of the network is as follows: the current hidden state is jointly determined by the current text parameter stream input, the previous hidden state, and the weighted aggregation of historical hidden states. The weighted aggregation of historical hidden states is achieved through an attention decay factor, which decreases as the time difference increases, emphasizing the importance of recent hidden states. Finally, the constraint framework is generated from the hidden state through the sigmoid activation function, and the output value is limited between 0 and 1, representing the low-frequency constraint conditions at the policy level for subsequent multi-granularity instruction fusion.

[0056] The attention decay factor refers to a dynamic weight function in the policy layer memory-augmented network that quantifies the influence degree of historical hidden states on the current hidden state. Specifically, the attention decay factor is calculated based on the time difference between the current time point and the historical time point, and its value decreases as the time difference increases, usually represented in an exponential decay form.

[0057] The policy layer memory-augmented network combines the attention mechanism to effectively capture the long-term dependence relationships in the text parameter stream and is suitable for generating the constraint framework at the policy level. The bidirectional recurrent structure allows the network to consider both past and future information in the text parameter stream, enhancing the understanding and prediction ability of policy instructions. The design of the attention decay factor enables the network to dynamically adjust the emphasis on historical information, enhancing the sensitivity to recent policy changes, ensuring that the constraint framework reflects both long-term trends and adapts to short-term adjustments. This processing method provides stable policy guidance for fusing the instruction stream, ensuring the coordination between tactical execution and policy goals.

[0058] S4.3, the spatio-temporal gating mechanism realizes multi-granularity instruction fusion: The multi-granularity instruction fusion of the millisecond-level action sequence and the constraint framework is realized through the spatio-temporal gating mechanism to generate a fused instruction stream. The spatio-temporal gating unit first calculates the dynamic fusion coefficient, which is jointly determined by the millisecond-level action sequence, the constraint framework, and their spatio-temporal correlation term. The spatio-temporal correlation term is obtained by integrating the product of the millisecond-level action sequence and the constraint framework within a local time window, reflecting their short-term cooperation degree. Subsequently, the millisecond-level action sequence and the constraint framework are respectively multiplied by a preset weight matrix, and then added with the spatio-temporal correlation term. The resulting result is input into the sigmoid function, and the output value is limited between 0 and 1. Finally, the fused instruction stream is the weighted average of the millisecond-level action sequence and the constraint framework, and the weights are dynamically adjusted by the dynamic fusion coefficient. The output fused instruction stream is used as the input for subsequent processing. For example, generate the fused instruction stream : , where Contributions of dynamic balance tactical actions and policy constraints is a millisecond-level action sequence The policy constraint framework is a dimensionless eigenvalue sequence.

[0059] The spatio-temporal gating mechanism achieves a flexible balance between the millisecond-level action sequence and the constraint framework through a dynamic fusion coefficient, and can adaptively adjust the contribution ratio of the two in different situations. The introduction of the spatio-temporal correlation term enhances the sensitivity of the fusion process to short-term synergy, ensuring that the fusion instruction stream can not only quickly respond to tactical requirements but also follow the constraint conditions at the policy level. This multi-granularity fusion method improves the coordination and consistency of the instructions, avoids conflicts between tactical actions and policy goals, and provides a stable and efficient input for subsequent non-linear oscillation monitoring and group policy convergence.

[0060] In step S4, the conflict-resolution visual parameter stream is converted into a millisecond-level action sequence through the tactical layer spiking neural network, the constraint framework is generated from the text parameter stream through the policy layer memory-enhanced network, and the spatio-temporal gating mechanism is used to achieve the dynamic fusion of the millisecond-level action sequence and the constraint framework, and finally the fusion instruction stream is output. While retaining the high-frequency response at the tactical level, the fusion instruction stream is modulated by the constraint framework at the policy level, ensuring the consistency between tactical execution and policy goals. This enables step S4 to efficiently process the conflict-resolution visual parameter stream and text parameter stream.

[0061] In step S4, the conflict-resolution visual parameters are processed by the tactical layer spiking neural network to generate a millisecond-level action sequence, the text parameters are processed by the policy layer memory-enhanced network to generate a constraint framework, and the spatio-temporal gating mechanism is used to fuse multi-granularity instructions to generate a fusion instruction stream. This fusion instruction stream achieves a dynamic balance between tactical actions and policy constraints and has the characteristics of high-frequency response and low-frequency guidance. However, due to the non-linear interaction of multi-modal parameters and the complexity of group collaborative decision-making, the fusion instruction stream may exhibit non-linear oscillation characteristics and even lead to policy deviation at the edge of chaos. To solve this problem, step S5 monitors the non-linear oscillation characteristics of the fusion instruction stream in real time, constructs a damping controller based on the Lyapunov stability theory, and injects a negative feedback correction signal when the edge of chaos is detected to ensure the convergence and stability of the group policy.

[0062] Step S5 includes the following: S5.1, Non-linear oscillation characteristic monitoring: For the fused instruction stream, the nonlinear oscillation characteristics of the fused instruction stream are monitored through phase space reconstruction technology. The fused instruction stream is the dimensionless eigenvalue sequence generated in the aforementioned step S4, which fuses the instruction information of tactical actions and policy constraints. The purpose of phase space reconstruction is to convert the time series of the fused instruction stream into a trajectory in a high-dimensional phase space to reveal the potential dynamic characteristics of the fused instruction stream. The specific processing process is as follows: First, appropriate embedding dimension and time delay parameters are selected. The embedding dimension represents the dimension of the phase space, and the time delay represents the time interval during reconstruction. Both are determined through techniques such as the mutual information method and the false nearest neighbor method to ensure that the reconstructed phase space can accurately reflect the dynamic behavior of the fused instruction stream. Then, based on the selected embedding dimension and time delay, a reconstructed vector is constructed. The reconstructed vector consists of the values of the fused instruction stream at different time lags, forming trajectory points in the phase space. Then, in the reconstructed phase space, the divergence characteristics of the trajectory are calculated. The specific operation is to track the deviation degree of adjacent trajectory points with time evolution and calculate the largest Lyapunov exponent. The calculation method of the largest Lyapunov exponent is to take the limit value of the logarithmic growth rate of the norm of the trajectory deviation vector with time evolution, representing the divergence speed of the trajectory in the phase space. If the largest Lyapunov exponent is positive, it is determined that the fused instruction stream has chaotic behavior; if the absolute value of the largest Lyapunov exponent is close to zero, it is determined that the fused instruction stream is at the edge of chaos, which may lead to unstable strategies.

[0063] For example: Input: Fused instruction stream , which is the dimensionless eigenvalue sequence generated in step S4 and represents the fusion result of multi-granularity instructions.

[0064] Processing: Use phase space reconstruction technology to analyze the dynamic characteristics. Select the embedding dimension and time delay , and construct the reconstructed vector:

[0065] Among them, and are determined through the mutual information method and the false nearest neighbor method.

[0066] Calculate the Lyapunov exponent spectrum in the reconstructed phase space to measure the divergence characteristics of the trajectory. The largest Lyapunov exponent is defined as:

[0067] Among them, represents the deviation vector between adjacent trajectories, represents the Euclidean norm.

[0068] According to the value of, judge the system state: If , the fused instruction stream has chaotic behavior; if is close to , it is at the edge of chaos.

[0069] Output: The maximum Lyapunov exponent , indicating the degree of chaos of the fused instruction stream.

[0070] The phase space reconstruction technique visualizes and quantifies the non-linear dynamic characteristics of the fused instruction stream, and is applicable to the analysis of chaotic and oscillatory behaviors in complex systems. The maximum Lyapunov exponent, as a quantitative index of chaotic behavior, can accurately judge whether the fused instruction stream is in an unstable state, providing a scientific basis for the subsequent construction of a damping controller and negative feedback injection. This monitoring method ensures real-time mastery of the dynamic characteristics of the fused instruction stream, overcomes the limitations of traditional linear analysis methods in dealing with non-linear systems, and improves the accuracy and reliability of monitoring.

[0071] S5.2, Damping controller construction: Based on the Lyapunov stability theory, a damping controller is constructed for the fused instruction stream to ensure the stability of the fused instruction stream. The design of the damping controller depends on the construction of the Lyapunov function and the realization of the stability conditions. The specific processing process is as follows: First, a Lyapunov function is defined, which takes half of the square of the current value of the fused instruction stream to represent the energy of the system. Then, the time derivative of the Lyapunov function is calculated, and the time derivative is the product of the current value of the fused instruction stream and its rate of change. To ensure the stability of the system, that is, the Lyapunov function decreases with time, a damping term is introduced, and a modified instruction stream is designed such that the rate of change of the modified instruction stream is negatively correlated with the fused instruction stream. Specifically, the dynamic evolution of the damping term is designed as a combination of the negative value of the fused instruction stream and the rate of change of the fused instruction stream to achieve negative feedback control of the fused instruction stream. By solving the dynamic evolution relationship of the damping term, the damping term is obtained as a combination of the negative integral term of the fused instruction stream and the fused instruction stream change adjustment term. Finally, the damping term is used as a negative feedback signal to dynamically adjust the fused instruction stream in subsequent steps.

[0072] For example: Input: The fused instruction stream , the maximum Lyapunov exponent .

[0073] Design a damping controller based on the Lyapunov stability theory, and define the Lyapunov function:

[0074] Its time derivative is:

[0075] To ensure stability (i.e., ), introduce the damping term , generate a corrected instruction stream:

[0076] and design the damping dynamics:

[0077] where, is the dimensionless damping coefficient, which controls the convergence rate.

[0078] Solve the dynamic equation of the damping term:

[0079] Integrate to obtain:

[0080] Output: the damping term , which is a dimensionless eigenvalue sequence representing the negative feedback correction signal.

[0081] Lyapunov stability theory is a classical method for controlling the stability of nonlinear systems. By constructing a Lyapunov function and designing the damping term, it ensures that the dynamic behavior of the fused instruction stream converges to a stable state. The design of the damping controller can adaptively adjust the evolution trend of the fused instruction stream, prevent the fused instruction stream from falling into chaos or divergence, and improve the robustness and reliability of group collaborative decision-making.

[0082] S5.3, Chaos edge detection and negative feedback injection: Based on the monitoring results of the maximum Lyapunov exponent, detect whether the fused instruction stream is at the edge of chaos and inject a negative feedback correction signal if necessary. The specific processing process is as follows: First, set a predefined chaos edge threshold. When the absolute value of the maximum Lyapunov exponent is less than the chaos edge threshold, it is determined that the fused instruction stream is at the edge of chaos and there is a risk of policy deviation. Then, construct the corrected instruction stream: If it is detected that the fused instruction stream is at the edge of chaos, then use the weighted sum of the fused instruction stream and the damping term as the corrected instruction stream; if the chaos edge is not detected, keep the fused instruction stream unchanged. The injection intensity of the damping term is controlled by a preset adjustment parameter to balance the correction effect and the original characteristics of the fused instruction stream. Finally, output the corrected instruction stream as the instruction stream after stability control for subsequent decision execution.

[0083] The chaos edge detection and negative feedback injection mechanism can intervene in a timely manner when the fused instruction stream is about to become unstable, ensuring the convergence and stability of the group strategy. By dynamically adjusting the corrected instruction stream and injecting the damping term only when necessary, it avoids excessive interference with the normal fused instruction stream and maintains the agility of tactical response. This adaptive control strategy improves the anti-interference ability and stability of the multi-modal decision-making system in adversarial scenarios and provides a reliable technical guarantee for group collaborative decision-making.

[0084] For example: Input: the maximum Lyapunov exponent , fusion instruction stream , damping term .

[0085] Processing: Set the threshold of the edge of chaos , when , it is determined that the fusion instruction stream is at the edge of chaos and a correction signal needs to be injected.

[0086] Construct the corrected instruction stream :

[0087] where is an adjustment parameter to control the damping injection intensity.

[0088] Output: the corrected instruction stream , which is a dimensionless eigenvalue sequence representing the instruction stream after stability control.

[0089] In step S5, the nonlinear oscillation characteristics of the fusion instruction stream are monitored through phase space reconstruction technology, a damping controller is constructed using Lyapunov stability theory, and a negative feedback correction signal is dynamically injected when the edge of chaos is detected, finally generating a corrected instruction stream. The corrected instruction stream effectively suppresses the strategy deviation in group collaborative decision-making, ensuring the stability and convergence of multimodal instruction fusion in complex environments. The monitoring of the nonlinear oscillation characteristics in the first sub-step provides a dynamic basis for subsequent control, the construction of the damping controller in the second sub-step lays the foundation for stability control, and the detection of the edge of chaos and negative feedback injection in the third sub-step achieve the precise implementation of dynamic correction. This design of step-by-step monitoring, control, and correction enables step S5 to efficiently process the dynamic characteristics of the fusion instruction stream, providing a solid guarantee for the robustness and efficiency of the entire decision-making system, and is applicable to group collaborative decision-making that requires high stability and reliability.

[0090] The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0091] It should be noted that the system of the present invention can be deployed on the device itself to achieve embedded applications, or can also run on a PC with a user interface or other terminals, so as to meet various hardware environments and usage requirements.

[0092] Only some exemplary embodiments of the present invention have been described by way of illustration. Undoubtedly, for those of ordinary skill in the art, the described embodiments can be modified in various different ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0093] It should be noted that in this text, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0094] As described above, this is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A decision-making method for existing parameter hybrid drive, characterized in that Including the steps: S1: Detecting the mutation of the fluctuation characteristics of the visual parameter flow and identifying the strategic instruction events of the text parameter flow, aligning the key conflict intervals of the multi-modal parameters through the dynamic time warping algorithm, and generating a set of conflict period coordinates with non-uniform distribution; S2: Within the defined conflict period coordinates, performing event-driven adaptive downsampling on the high-frequency visual parameters to align their time bases with the text parameters, and using time-frequency analysis techniques to extract the phase compensation vector to eliminate the inter-modal phase difference; S3: Jointly analyzing the time-frequency energy distribution and causal correlation intensity of the synchronized parameters, constructing a counterfactual intervention model to generate a frequency band suppression weight matrix, directionally weakening the energy of the frequency band causing strategic contradictions and strengthening the strategic causal path; S4: Inputting the conflict-free visual parameters into the tactical layer pulse neural network to generate a millisecond-level action sequence, and inputting the text parameters into the strategic layer memory-enhanced network to generate a constraint framework, and realizing multi-granularity instruction fusion through the spatio-temporal gating mechanism; S5: Real-time monitoring the non-linear oscillation characteristics of the fused instruction flow, constructing a damping controller based on the Lyapunov stability theory, and injecting a negative feedback correction signal to maintain the convergence of the group strategy when detecting the characteristics of the edge of chaos.

2. The decision-making method for hybrid drive based on existing parameters according to claim 1, wherein Step S1 includes the following: First, detecting the mutation points of the fluctuation characteristics by calculating the local fluctuation entropy of the visual parameter flow, statistically calculating the probability distribution of the change amplitude of the visual parameter flow in each time window and calculating the information entropy, and marking the central time point of the corresponding time window as the visual mutation point when the local fluctuation entropy exceeds the preset entropy threshold; Then, identifying the strategic instruction events of the text parameter flow by using the exponential smoothing cumulative deviation method, applying exponential smoothing to the text parameter flow to reduce short-term fluctuations, calculating the dynamic deviation between the smoothed parameter and the moving average value, and marking the corresponding time point as the strategic instruction event point when the dynamic deviation exceeds the preset deviation threshold; Next, using the dynamic time warping algorithm to perform temporal alignment on the set of visual mutation points and the set of strategic instruction event points, constructing a cumulative distance matrix and solving the minimum cumulative distance path through dynamic programming to generate a set of aligned time point pairs; Finally, generating a set of conflict period coordinates based on the set of aligned time point pairs, defining the conflict period for each pair of aligned time points and performing expansion and merging to form a set of conflict period coordinates with non-uniform distribution.

3. The decision-making method for existing parameter hybrid drive according to claim 2, wherein Step S2 includes the following: Within each conflict period, first calculating the local change rate of the high-frequency visual parameter flow to identify the event trigger points, and marking the corresponding time point as the event trigger point when it is determined that the local change rate exceeds the preset change rate threshold; Then, performing adaptive downsampling based on the event trigger points, retaining the parameter values at the event trigger points, and using linear interpolation for non-event trigger points to generate approximate values, so that the number of sampling points of the downsampled high-frequency visual parameter flow within the conflict period is the same as the number of sampling points of the low-frequency text parameter flow within the same conflict period, realizing the alignment of time bases.

4. The decision-making method for hybrid drive oriented to existing parameters according to claim 3, characterized in that Step S2 also includes the following: Next, performing continuous wavelet transform on the downsampled high-frequency visual parameter flow and low-frequency text parameter flow respectively to generate their time-frequency spectra, and calculating the phase difference between the two at each time point and scale; Construct a phase compensation vector based on the phase difference, and take the negative value of the phase difference at the dominant scale of each time point as the compensation phase; Finally, apply phase rotation adjustment to the downsampled high-frequency visual parameter stream, multiply the complex signal of each time point by the complex exponential term, where the phase angle of the complex exponential term is the compensation phase value, and take the real part to generate the synchronized parameter stream, so as to achieve alignment in time and phase with the low-frequency text parameter stream.

5. The decision-making method for hybrid drive oriented to existing parameters according to claim 4, characterized in that Step S3 includes the following: Perform wavelet packet decomposition on the synchronized visual parameter stream to generate a time-frequency energy matrix, and perform short-time Fourier transform on the text parameter stream to generate a main frequency energy feature matrix; Calculate the cross-coherence matrix based on the time-frequency energy matrix of the visual parameter stream and the time-frequency energy matrix of the text parameter stream to quantify the energy coupling strength of the physical layer; Construct a structural equation model to analyze the causal effect of the visual parameter stream on the text parameter stream, and calculate the conditional entropy change value through counterfactual reasoning to characterize the causal association strength of the logic layer; Fuse the cross-coherence matrix, the sign of the causal strength coefficient, and the conditional entropy change value to generate a dynamic suppression weight matrix, apply suppression to the frequency bands with high coherence and negative causal association through an exponential decay function, and apply positive compensation to the policy main frequency at the same time; Finally, perform time-frequency domain filtering on the synchronized visual parameter stream to generate a conflict-free visual parameter stream, so as to weaken the energy of the frequency bands that cause policy contradictions and strengthen the policy causal path.

6. The decision-making method for hybrid drive based on existing parameters according to claim 5, wherein Step S4 includes the following: Perform dynamic threshold modulation on the conflict-free visual parameter stream through a tactical layer spiking neural network, use a non-linear neuron model to simulate the process of generating tactical actions, where the neuron membrane potential evolves over time driven by the product of the conflict-free visual parameter stream and the phase modulation term, and the dynamic threshold is adjusted according to the local energy accumulation of the conflict-free visual parameter stream. When the neuron membrane potential exceeds the dynamic threshold, a spike signal is triggered and the membrane potential is reset, and the spike sequence is mapped to a millisecond-level action sequence through a multi-layer structure.

7. The decision-making method for existing parameter hybrid drive according to claim 6, characterized in that, Step S4 includes the following: Perform attention-enhanced bidirectional cyclic processing on the text parameter stream through a policy layer memory-enhanced network. The update of the hidden state is jointly determined by the current text parameter stream input, the previous moment's hidden state, and the weighted summary of the historical hidden states. The weighted summary of the historical hidden states is calculated through an attention attenuation factor, and finally a constraint framework is generated through a sigmoid activation function.

8. The decision-making method for hybrid drive oriented to existing parameters according to claim 7, characterized in that Step S4 includes the following: Realize the dynamic fusion of the millisecond-level action sequence and the constraint framework through a spatio-temporal gating mechanism, construct a spatio-temporal gating unit to calculate the dynamic fusion coefficient, where the dynamic fusion coefficient is determined by the millisecond-level action sequence, the constraint framework, and their spatio-temporal correlation term. The spatio-temporal correlation term is calculated through the product integration of the millisecond-level action sequence and the constraint framework within a local time window. Finally, the fused instruction stream is generated by the weighted average of the millisecond-level action sequence and the constraint framework.

9. The decision-making method for hybrid drive oriented to existing parameters according to claim 8, characterized in that, Step S5 includes the following: Perform dynamic characteristic analysis on the fused instruction stream through phase space reconstruction technology, select the embedding dimension and time delay parameters to construct a reconstruction vector, and calculate the maximum Lyapunov exponent in the reconstructed phase space to judge the chaotic behavior of the fused instruction stream; Design a damping controller through Lyapunov stability theory, define the Lyapunov function and design the dynamic evolution of the damping term to ensure the stability of the fused instruction stream; Monitor the maximum Lyapunov exponent through the edge-of-chaos detection mechanism. When the absolute value of it is less than the edge-of-chaos threshold, it is determined that the fused instruction stream is at the edge of chaos, and a damping term is injected as a negative feedback correction signal to generate a corrected instruction stream to maintain the convergence of the swarm strategy.

Images