A Multi-Agent Privacy Protection Analysis Method and System Based on Pulse Communication Mechanism

Abstract

This invention relates to the field of computer security technology, and in particular to a method and system for privacy protection analysis of multi-agent communication mechanisms using pulse communication. The method includes the following steps: acquiring a continuous input signal sequence corresponding to a multi-agent system and performing spatiotemporal sparse pulse transformation to generate a multi-agent spatiotemporal sparse pulse sequence; constructing a hierarchical spatiotemporal pulse neural network and combining it with a pulse network training method based on differential privacy to perform pulse communication decision output and pulse perturbation randomization processing based on a combination of pulse time jitter and pulse deletion to generate a multi-agent pulse perturbation randomized sequence; constructing privacy leakage attack scenario models corresponding to pulse sequence reconstruction attacks, member reasoning attacks, and attribute reasoning attacks, and performing privacy leakage attack simulation and privacy leakage protection analysis to generate a privacy-efficiency protection optimization strategy for multi-agent pulse communication. This invention can achieve a balanced optimization of privacy protection and communication efficiency in multi-agent systems.

Description

Technical Field

[0001] This invention relates to the field of computer security technology, and in particular to a multi-agent privacy protection analysis method and system based on pulse communication mechanism. Background Technology

[0002] With the rapid development of intelligent technologies, multi-agent systems (MAS) have been widely applied in many fields, such as intelligent transportation, intelligent manufacturing, robot collaboration, and distributed energy management. These MAS achieve efficient task completion through mutual cooperation and information sharing. Pulse communication mechanism is a communication method that mimics the signal transmission mechanism of biological nervous systems. In this mechanism, information is transmitted through the triggering and propagation of pulse signals, rather than the traditional data packet exchange method. This communication mechanism has high time efficiency and low latency, making it suitable for multi-agent systems requiring rapid response and high-frequency data exchange. However, current traditional privacy protection methods mainly rely on encryption technologies, homomorphic encryption, and privacy computing. But these methods still face a key problem when dealing with high-frequency, real-time pulse communication mechanisms: how to ensure the privacy of agents is not leaked while maintaining information transmission efficiency. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide a multi-agent privacy protection analysis method and system using a pulse communication mechanism to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a multi-agent privacy protection analysis method based on a pulse communication mechanism is provided, comprising the following steps:

[0005] Step S1: Obtain the continuous input signal sequence corresponding to the multi-agent system, and perform spatiotemporal sparse pulse transformation on the continuous input signal sequence corresponding to the multi-agent system based on the sparse matrix generation method of time coding and rate coding to generate a multi-agent spatiotemporal sparse pulse sequence.

[0006] Step S2: Construct a hierarchical spatiotemporal spiking neural network, and input the multi-agent spatiotemporal sparse spiking sequence into the hierarchical spatiotemporal spiking neural network. Combine the spiking network training method based on differential privacy to perform spiking communication decision output and generate multi-agent time-series key communication spiking sequences.

[0007] Step S3: Perform pulse perturbation randomization processing on the multi-agent time-series key communication pulse sequence based on a combination of pulse time jitter and pulse deletion to generate a multi-agent pulse perturbation randomization sequence;

[0008] Step S4: Construct privacy leakage attack scenario models corresponding to pulse sequence reconstruction attacks, member reasoning attacks, and attribute reasoning attacks. Use Monte Carlo simulation combined with the corresponding privacy leakage attack scenario models to simulate privacy leakage attacks on multi-agent pulse perturbation randomized sequences to generate multi-agent pulse privacy leakage probabilities under different attack scenarios. Based on the multi-agent pulse privacy leakage probabilities under different attack scenarios, perform privacy leakage protection analysis on multi-agent pulse perturbation randomized sequences to generate privacy-efficiency protection optimization strategies for multi-agent pulse communication.

[0009] Furthermore, step S1 includes the following steps:

[0010] Step S11: Obtain the continuous input signal sequence corresponding to the multi-agent system;

[0011] Step S12: Generate input signal pulses from the continuous input signal sequence corresponding to the multi-agent system to generate a multi-agent input signal pulse sequence;

[0012] Step S13: The time-coded sparse matrix generation method performs time-delay sparse matrix transformation on the multi-agent input signal pulse sequence to generate a multi-agent signal sparse matrix with time-delay characteristics;

[0013] Step S14: The sparse matrix generation method based on rate coding performs pulse rate sparse matrix transformation on the multi-agent input signal pulse sequence to generate a multi-agent signal sparse matrix with rate characteristics.

[0014] Step S15: Perform spatiotemporal sparse pulse transformation on the multi-agent input signal pulse sequence based on the multi-agent signal sparse matrix with time delay and rate characteristics to generate a multi-agent spatiotemporal sparse pulse sequence.

[0015] Furthermore, step S13 includes the following steps:

[0016] Step S131: Obtain the corresponding signal pulse time interval and signal pulse waveform characteristics by inputting the signal pulse sequence through multiple agents;

[0017] Step S132: Determine the dilution time window based on the signal pulse time interval and signal pulse waveform characteristics to obtain the size of the sparse signal pulse time window related to the original signal time distribution of the multi-agent agent;

[0018] Step S133: Based on the size of the sparse time window of the signal pulse related to the time distribution of the original signal of the multi-agent agent, perform pulse amplitude and time sequence position statistics for each sparse time window moment in the input signal pulse sequence of the multi-agent agent to obtain the multi-agent agent signal pulse amplitude and multi-agent agent signal time sequence position corresponding to each sparse time window moment.

[0019] Step S134: Calculate the time delay distribution difference at different times based on the amplitude of the multi-agent signal pulse and the timing position of the multi-agent signal corresponding to each sparse time window, and obtain the time delay distribution difference between the multi-agent signal pulses at different sparse time windows.

[0020] Step S135: Based on the time delay distribution differences between multi-agent signal pulses at different sparse time windows and combined with the time-coded sparse matrix generation method, the multi-agent input signal pulse sequence is transformed into a time delay sparse matrix to generate a multi-agent signal sparse matrix with time delay characteristics.

[0021] Furthermore, step S14 includes the following steps:

[0022] Step S141: Perform statistical calculations on the pulse frequency and amplitude within each signal pulse window of the multi-agent input signal pulse sequence to obtain the corresponding multi-agent signal pulse frequency and multi-agent signal pulse amplitude within each signal pulse window;

[0023] Step S142: Obtain the corresponding multi-agent pulse signal intensity through the multi-agent input signal pulse sequence, and perform pulse rate modulation on the multi-agent input signal pulse sequence based on the multi-agent pulse signal intensity to generate the modulation signal pulse rate corresponding to the multi-agent signal pulse.

[0024] Step S143: Based on the modulation signal pulse rate corresponding to the multi-agent signal pulse and combined with the sparse matrix generation method of rate coding, the pulse rate sparse weighted reconstruction of the multi-agent signal pulse frequency and the multi-agent signal pulse amplitude corresponding to each signal pulse window is performed to generate a multi-agent signal sparse matrix with rate characteristics.

[0025] Furthermore, step S15 includes the following steps:

[0026] Step S151: Based on the sparse matrix of multi-agent signals with time delay characteristics, perform time delay sparsification filtering on the pulse sequence of multi-agent input signals to obtain the multi-agent time delay sparse filtered pulse sequence.

[0027] Step S152: Based on the sparse matrix of the multi-agent signal with rate characteristics, the multi-agent time delay sparse filter pulse sequence is rate-sparsely modulated in spatial location to generate a multi-agent rate sparse modulation pulse sequence.

[0028] Step S153: Obtain the pulse energy magnitude corresponding to each pulse signal through the multi-agent rate sparse modulation pulse sequence, and evaluate the pulse attenuation based on the pulse energy magnitude corresponding to each pulse signal to obtain the pulse attenuation factor corresponding to the multi-agent pulse signal.

[0029] Step S154: Based on the pulse attenuation factor corresponding to the multi-agent pulse signal, optimize the spatiotemporal pulse distribution of each pulse signal in the multi-agent rate sparse modulation pulse sequence to generate a multi-agent spatiotemporal sparse pulse sequence.

[0030] Furthermore, step S2 includes the following steps:

[0031] Step S21: Construct a corresponding hierarchical spatiotemporal spiking neural network by adopting a structure that combines recurrent neural networks and long short-term memory networks. Specifically, it includes a bottom layer, a middle layer, and a top layer. The bottom layer is used to process the original spiking sequence, the middle layer is used to extract the corresponding spiking temporal features, and the top layer is used to make decisions and output the corresponding communication spiking sequence.

[0032] Step S22: Input the multi-agent spatiotemporal sparse pulse sequence into the bottom layer of the hierarchical spatiotemporal spiking neural network for pulse preprocessing, and input the preprocessed multi-agent spatiotemporal sparse pulse sequence into the middle layer of the hierarchical spatiotemporal spiking neural network to train the pulse extraction network by introducing a pulse temporal attention mechanism. At the same time, during the training process, a pulse network training method based on differential privacy is designed and constructed to introduce Gaussian noise into the spiking neural network and prevent gradient inversion attacks by combining the gradient pruning strategy in the training process, so as to train and generate the pulse temporal feature output corresponding to each agent.

[0033] Step S23: Calculate the pulse interaction frequency and pulse sensitivity between agents by using the pulse timing features output from the top layer of the hierarchical spatiotemporal spiking neural network. Then, dynamically configure the corresponding privacy budget based on the pulse interaction frequency and pulse sensitivity between agents. Parameters, while based on this privacy budget The parameters are used to make decisions at the top level, resulting in the corresponding communication pulse sequence, which generates the multi-agent time-series key communication pulse sequence.

[0034] Furthermore, step S3 includes the following steps:

[0035] Step S31: Perform pulse timing perturbation analysis on the multi-agent time-series key communication pulse sequences to obtain the distribution characteristics of multi-agent communication pulse timing perturbation;

[0036] Step S32: Obtain the pulse communication disturbance delay and pulse communication disturbance frequency difference between each agent through the pulse timing disturbance distribution characteristics of multi-agent communication, and perform pulse time jitter amplitude statistics on each pulse signal in the multi-agent timing key communication pulse sequence based on the pulse communication disturbance delay and pulse communication disturbance frequency difference between each agent to obtain the communication pulse signal time jitter amplitude between each agent.

[0037] Step S33: Generate a corresponding multi-agent pulse time jitter amplitude map based on the time jitter amplitude of the communication pulse signals between each agent, and perform pulse selective deletion optimization on each pulse signal in the multi-agent time-series key communication pulse sequence based on the multi-agent pulse time jitter amplitude map, so as to filter and delete redundant pulse signals with consistent time jitter amplitude of communication pulse signals, so as to generate a multi-agent pulse optimization deletion sequence.

[0038] Step S34: Perform pulse perturbation distribution randomization processing on the timing of the corresponding remaining pulse signals in the multi-agent pulse optimization deletion sequence to generate a multi-agent pulse perturbation randomization sequence.

[0039] Furthermore, step S4 includes the following steps:

[0040] Step S41: Construct privacy leakage attack scenario models corresponding to pulse sequence reconstruction attack, membership reasoning attack, and attribute reasoning attack;

[0041] Step S42: Use Monte Carlo simulation combined with the corresponding privacy leakage attack scenario model to simulate privacy leakage attacks on the randomized sequence of multi-agent pulse perturbation, so as to generate the corresponding multi-agent pulse privacy leakage probability under different attack scenarios.

[0042] Step S43: Based on the multi-agent pulse privacy leakage probability under different attack scenarios, perform privacy leakage diffusion impact analysis on the multi-agent pulse perturbation randomization sequence to obtain the diffusion path and impact range of the corresponding pulse privacy leakage event in the multi-agent system;

[0043] Step S44: Based on the propagation path and impact range of the corresponding impulse privacy leakage event in the multi-agent system, construct a privacy-efficiency optimization objective corresponding to the privacy protection level, communication bandwidth consumption, and computational latency;

[0044] Step S45: Based on the privacy-efficiency optimization objective corresponding to the privacy protection level, communication bandwidth consumption and computation delay, perform privacy leakage protection analysis on the randomized sequence of multi-agent pulse perturbation. Use the NSGA-II algorithm to solve for the Pareto optimal solution set corresponding to the privacy-efficiency optimization objective. Based on the Pareto optimal solution set, obtain the balance strategy between privacy and efficiency for pulse signal communication to generate the privacy-efficiency protection optimization strategy for multi-agent pulse communication.

[0045] Furthermore, step S42 includes the following steps:

[0046] Monte Carlo simulation was used in conjunction with privacy-leakage attack scenario models corresponding to pulse sequence reconstruction attack, membership reasoning attack and attribute reasoning attack to simulate and infer pulse attack on randomized sequences of multi-agent pulse perturbation, so as to generate the multi-agent pulse perturbation attack generation process under different attack scenarios.

[0047] The pulse decision-making process of multi-agent pulse perturbation attack generation under different attack scenarios is subjected to multi-round reverse reasoning to generate multi-agent pulse perturbation attack decision trees under different attack scenarios.

[0048] By obtaining the multi-agent pulse perturbation stagnation time points in different attack scenarios through the multi-agent pulse perturbation attack decision tree corresponding to different attack scenarios, and performing leakage probability prediction calculation on the multi-agent pulse perturbation attack decision tree corresponding to different attack scenarios based on the multi-agent pulse perturbation stagnation time points in different attack scenarios, the multi-agent pulse privacy leakage probability corresponding to different attack scenarios is generated.

[0049] Furthermore, the present invention also provides a multi-agent privacy protection analysis system based on a pulse communication mechanism, for executing the multi-agent privacy protection analysis method based on the pulse communication mechanism described above. The multi-agent privacy protection analysis system based on the pulse communication mechanism includes:

[0050] The spatiotemporal sparse pulse conversion module is used to acquire the continuous input signal sequence corresponding to the multi-agent system, and perform spatiotemporal sparse pulse conversion on the continuous input signal sequence corresponding to the multi-agent system based on the sparse matrix generation method of time coding and rate coding to generate a multi-agent spatiotemporal sparse pulse sequence.

[0051] The pulse communication decision output module is used to construct a hierarchical spatiotemporal spiking neural network and input the multi-agent spatiotemporal sparse pulse sequence into the hierarchical spatiotemporal spiking neural network. Combined with the pulse network training method based on differential privacy, the pulse communication decision output is performed to generate the multi-agent time-series key communication pulse sequence.

[0052] The pulse perturbation randomization module is used to perform pulse perturbation randomization processing on the multi-agent time-critical communication pulse sequence based on a combination of pulse time jitter and pulse deletion, thereby generating a multi-agent pulse perturbation randomized sequence.

[0053] The pulse privacy protection analysis module is used to construct privacy leakage attack scenario models corresponding to pulse sequence reconstruction attacks, membership inference attacks, and attribute inference attacks. Monte Carlo simulation combined with the corresponding privacy leakage attack scenario models is used to simulate privacy leakage attacks on multi-agent pulse perturbation randomized sequences to generate multi-agent pulse privacy leakage probabilities under different attack scenarios. Based on the multi-agent pulse privacy leakage probabilities under different attack scenarios, privacy leakage protection analysis is performed on multi-agent pulse perturbation randomized sequences to generate privacy-efficiency protection optimization strategies for multi-agent pulse communication.

[0054] The beneficial effects of this invention are:

[0055] 1. The multi-agent privacy protection analysis method for pulse communication mechanism proposed in this invention, compared with the prior art, has the following advantages: It obtains the continuous input signal sequence corresponding to the multi-agent system and performs spatiotemporal sparse pulse conversion on these signals using a sparse matrix generation method based on time coding and rate coding. The core advantage of this step is that spatiotemporal sparse coding effectively converts the continuous input signal into a sparse pulse sequence, reducing the amount of data transmission and improving the efficiency of information processing. The sparse coding method can effectively compress the signal, reduce the computational burden of the system, and ensure that key information is not lost. Especially in multi-agent systems, the number of agents is very large, and the complexity of the signal also increases accordingly. Through this spatiotemporal sparse pulse conversion, unnecessary computational overhead can be greatly reduced, and the communication efficiency between multiple agents can be optimized. The generation of spatiotemporal sparse pulse sequences also provides more efficient and concise data input for subsequent spiking neural networks, laying a solid foundation for improving the overall system performance. Secondly, by constructing a hierarchical spatiotemporal spiking neural network, the signal sequence transformed by spatiotemporal sparse pulses is input into the network. Combined with a differential privacy-based spiking network training method, pulse communication decision output is generated, thus producing a multi-agent time-series critical communication pulse sequence. This fully exploits the spatiotemporal characteristics of the input signal, decomposing and processing the complex input signal at different levels, thereby improving the overall processing capability and accuracy of the multi-agent system. Through the hierarchical structure design, the neural network can capture key features of the signal at multiple scales, ensuring more efficient and accurate information transmission in the multi-agent system. The introduction of differential privacy effectively enhances data protection throughout the spiking network training process, avoiding the risk of data leakage. Privacy protection is a crucial issue in multi-agent systems, and differential privacy technology provides a feasible solution. By encrypting the training process, it ensures normal operation and decision output while maintaining data privacy. This privacy-preserving training method provides higher security and reliability for multi-agent system applications.Then, by performing pulse perturbation randomization processing on the multi-agent time-critical communication pulse sequence based on a combination of pulse time jitter and pulse deletion, the main advantage of this process is that by introducing perturbation randomization, the privacy protection effect can be further improved while maintaining the efficiency and stability of the system. Pulse time jitter and pulse deletion are two effective privacy protection methods that can disrupt the attacker's analysis of data without significantly affecting system performance, thereby effectively preventing potential attacks. Pulse time jitter makes it difficult for attackers to accurately infer the temporal relationship of the signal by randomly changing the time interval of the pulse, while pulse deletion further increases the difficulty for attackers to obtain the complete signal by deleting part of the pulse information. In this way, while ensuring the efficiency of information transmission, the privacy between agents is not leaked, thereby ensuring the privacy security and functional stability of the multi-agent system. Finally, privacy-leakage attack scenario models were constructed for pulse sequence reconstruction attacks, membership inference attacks, and attribute inference attacks. Monte Carlo simulations were then used to simulate privacy-leakage attacks on multi-agent pulse perturbation randomized sequences. The core advantage of this process is that by simulating various attack scenarios, the system's privacy protection capabilities under different attack conditions can be comprehensively evaluated. Monte Carlo simulations can verify the system's robustness in numerous experiments and identify potential weaknesses. Privacy-leakage attack simulations not only help designers understand the effectiveness of current protection measures but also provide important references for subsequent optimization. Based on the simulations, targeted privacy-leakage protection analysis can be performed on pulse perturbation randomized sequences, ultimately generating optimization strategies. These strategies will balance the relationship between privacy protection and system efficiency, providing the best privacy protection and performance balance solution. This privacy-efficiency protection optimization strategy can ensure that multi-agent systems maintain a high level of privacy security when facing complex attacks, while also guaranteeing the system's high efficiency in practical applications.

[0056] 2. The multi-agent privacy protection analysis system based on the pulse communication mechanism proposed in this invention is composed of a spatiotemporal sparse pulse conversion module, a pulse communication decision output module, a pulse perturbation randomization module, and a pulse privacy protection analysis module. It can realize the multi-agent privacy protection analysis method for any pulse communication mechanism described in this invention. It is used to combine the operations between the computer programs running on each module to realize the multi-agent privacy protection analysis method for the pulse communication mechanism. The internal structure of the system cooperates with each other, which can greatly reduce repetitive work and manpower input, and can quickly and effectively provide a more accurate and efficient multi-agent privacy protection analysis process for the pulse communication mechanism, thereby simplifying the operation process of the multi-agent privacy protection analysis system based on the pulse communication mechanism. Attached Figure Description

[0057] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0058] Figure 1 This is a flowchart illustrating the steps of the multi-agent privacy protection analysis method based on the pulse communication mechanism of the present invention.

[0059] Figure 2 for Figure 1 A detailed flowchart of step S1;

[0060] Figure 3 for Figure 2 A detailed flowchart of step S13. Detailed Implementation

[0061] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0062] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0063] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0064] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a multi-agent privacy protection analysis method using a pulse communication mechanism, the method comprising the following steps:

[0065] Step S1: Obtain the continuous input signal sequence corresponding to the multi-agent system, and perform spatiotemporal sparse pulse transformation on the continuous input signal sequence corresponding to the multi-agent system based on the sparse matrix generation method of time coding and rate coding to generate a multi-agent spatiotemporal sparse pulse sequence.

[0066] Step S2: Construct a hierarchical spatiotemporal spiking neural network, and input the multi-agent spatiotemporal sparse spiking sequence into the hierarchical spatiotemporal spiking neural network. Combine the spiking network training method based on differential privacy to perform spiking communication decision output and generate multi-agent time-series key communication spiking sequences.

[0067] Step S3: Perform pulse perturbation randomization processing on the multi-agent time-series key communication pulse sequence based on a combination of pulse time jitter and pulse deletion to generate a multi-agent pulse perturbation randomization sequence;

[0068] Step S4: Construct privacy leakage attack scenario models corresponding to pulse sequence reconstruction attacks, member reasoning attacks, and attribute reasoning attacks. Use Monte Carlo simulation combined with the corresponding privacy leakage attack scenario models to simulate privacy leakage attacks on multi-agent pulse perturbation randomized sequences to generate multi-agent pulse privacy leakage probabilities under different attack scenarios. Based on the multi-agent pulse privacy leakage probabilities under different attack scenarios, perform privacy leakage protection analysis on multi-agent pulse perturbation randomized sequences to generate privacy-efficiency protection optimization strategies for multi-agent pulse communication.

[0069] In the embodiments of this invention, please refer to Figure 1 The diagram shown illustrates the steps of the multi-agent privacy protection analysis method based on the pulse communication mechanism of the present invention. In this example, the multi-agent privacy protection analysis method based on the pulse communication mechanism includes the following steps:

[0070] Step S1: Obtain the continuous input signal sequence corresponding to the multi-agent system, and perform spatiotemporal sparse pulse transformation on the continuous input signal sequence corresponding to the multi-agent system based on the sparse matrix generation method of time coding and rate coding to generate a multi-agent spatiotemporal sparse pulse sequence.

[0071] In this embodiment of the invention, in an industrial IoT multi-agent system composed of six agents, data from each agent is collected in real time through a sensor network to obtain a continuous input signal sequence corresponding to the multi-agent system. Taking agent A as an example, its vibration sensor collects equipment vibration data at a sampling frequency of 2000Hz, obtaining 2000 continuous vibration amplitude data within 1 second. For example, the amplitudes of the first to fifth sampling points are 12mV, 15mV, 13mV, 14mV, and 16mV, respectively. The other five agents also collect temperature, pressure, and other data at the same or different frequencies through their respective sensors, which together constitute the continuous input signal sequence of the multi-agent system. Spatiotemporal sparse pulse conversion is performed based on a sparse matrix generation method using time coding and rate coding. In terms of time coding, the time axis is divided into 100-millisecond time windows. For the vibration signal sequence of agent A, if the amplitudes of the first to fifth sampling points are 12mV, 15mV, 13mV, 14mV, and 16mV, respectively, the amplitudes of the first to fifth sampling points are 12mV, 15mV, 13mV, 14mV, and 16mV, respectively, the amplitudes of the first five sampling points are 12mV, 15mV, 13mV, 14mV, and 16mV, respectively, the amplitudes of the first five sampling points are 16mV, 14mV, and 16mV, respectively, the amplitudes of the first five sampling points are 16mV, 14mV, 16mV, and 16mV, respectively, the amplitudes of the first five sampling points are 16mV, 14mV, 16mV, and 16mV, respectively, the amplitudes of the first five sampling points are 16mV, 14mV, 16mV, and 16mV, Within a window, if the maximum signal amplitude is 20mV, exceeding the set threshold of 18mV, then a pulse is determined to exist within that time window. Regarding rate encoding, the number of pulses generated by each agent within one second is counted as the pulse rate. Agent A generated 8 pulses within the aforementioned one second, with a pulse rate of 8 pulses / second. A spatiotemporal sparse matrix of 6 rows (corresponding to 6 agents) and 10 columns (corresponding to 10 time windows) is constructed. For each agent in each time window, if a pulse exists, the relevant parameters of that pulse (such as amplitude, rate, etc., after normalization) are filled in; if no pulse exists, 0 is filled in. For example, agent A has a pulse in the 3rd time window, with a normalized amplitude of 0.8 and a normalized rate of 0.7. The corresponding position in the matrix is ​​filled with a vector [0.8, 0.7] composed of these two values. By processing the continuous input signal sequences of all agents, a multi-agent spatiotemporal sparse pulse sequence is finally generated.

[0072] Step S2: Construct a hierarchical spatiotemporal spiking neural network, and input the multi-agent spatiotemporal sparse spiking sequence into the hierarchical spatiotemporal spiking neural network. Combine the spiking network training method based on differential privacy to perform spiking communication decision output and generate multi-agent time-series key communication spiking sequences.

[0073] In this embodiment of the invention, a hierarchical spatiotemporal spiking neural network is constructed. This network comprises a bottom layer, a middle layer, and a top layer. The bottom layer consists of 12 neurons, each corresponding to a feature dimension of the agent's signal, and receives a multi-agent spatiotemporal sparse pulse sequence as input. The middle layer has 8 LSTM neurons for extracting pulse temporal features. The top layer has 4 neurons responsible for outputting pulse communication decisions, inputting the multi-agent spatiotemporal sparse pulse sequence into the hierarchical spatiotemporal spiking neural network. In the bottom layer, the neurons perform weighted summation on the input pulses. The weights are pre-determined based on historical data using the least squares method. For example, for a certain pulse input from agent B, neuron 1 has a weight of 0.3, the pulse value is 0.6, and the weighted sum is 0.3. 0.6 = 0.18. The processed signal is transmitted to the middle layer. The LSTM neurons in the middle layer introduce a pulse-time attention mechanism. The formula for calculating the attention weight is... ,in , The current hidden state. For the first The characteristic vector of each pulse , , The weight matrix is ​​trainable. During training, a differential privacy-based spiking network training method is used, adding Gaussian noise to the gradient of each training iteration, with a noise standard deviation of [missing information]. ,in Sensitivity to gradients, = 0.5 is the privacy budget parameter. At the same time, a gradient pruning strategy is used to prune the gradient norm to C=0.8. After 200 training epochs, the network converges and finally outputs the multi-agent temporal key communication pulse sequence at the top-level decision.

[0074] Step S3: Perform pulse perturbation randomization processing on the multi-agent time-series key communication pulse sequence based on a combination of pulse time jitter and pulse deletion to generate a multi-agent pulse perturbation randomization sequence;

[0075] In this embodiment of the invention, a pulse perturbation randomization process based on a combination of pulse time jitter and pulse deletion is applied to the multi-agent time-critical communication pulse sequence. The pulse time jitter range is set to ±50 milliseconds, and the pulse deletion probability is 0.2. Taking a pulse from agent C as an example, its original generation time is 0.5 seconds. Within the jitter range, a random time offset is generated, assuming it is 20 milliseconds, so the new generation time of the pulse becomes 0.5 + 0.02 = 0.52 seconds. For each pulse, a random number between 0 and 1 is generated and compared with the pulse deletion probability. If the random number is less than 0.2, the pulse is deleted. Assuming the generated random number is 0.15, which is less than 0.2, the pulse is deleted. This operation is performed on all pulses in the multi-agent time-critical communication pulse sequence. For example, if the original sequence contains 100 pulses, after processing, 20 pulses are deleted, and the remaining pulses undergo time jitter adjustment, ultimately generating a multi-agent pulse perturbation randomized sequence.

[0076] Step S4: Construct privacy leakage attack scenario models corresponding to pulse sequence reconstruction attacks, member reasoning attacks, and attribute reasoning attacks. Use Monte Carlo simulation combined with the corresponding privacy leakage attack scenario models to simulate privacy leakage attacks on multi-agent pulse perturbation randomized sequences to generate multi-agent pulse privacy leakage probabilities under different attack scenarios. Based on the multi-agent pulse privacy leakage probabilities under different attack scenarios, perform privacy leakage protection analysis on multi-agent pulse perturbation randomized sequences to generate privacy-efficiency protection optimization strategies for multi-agent pulse communication.

[0077] In this embodiment of the invention, privacy leakage attack scenario models corresponding to pulse sequence reconstruction attacks, membership inference attacks, and attribute inference attacks are constructed. In the pulse sequence reconstruction attack scenario model, the attacker uses known pulse generation rules and communication protocols, employing a least squares-based reconstruction algorithm to attempt to reconstruct the original pulse sequence from the multi-agent pulse perturbation randomized sequence. In the membership inference attack scenario model, the attacker uses a support vector machine algorithm to determine whether a certain agent belongs to the system based on the historical communication data and current observation sequence of the multi-agent system. In the attribute inference attack scenario model, the attacker constructs a random forest classification model based on the statistical characteristics of the agent's pulse sequence to infer the agent's attributes. Monte Carlo simulation combined with the attack scenario models is used to simulate privacy leakage attacks on the multi-agent pulse perturbation randomized sequence, with the number of simulations set to 10,000. In the pulse sequence reconstruction attack simulation, the mean square error between the reconstructed sequence and the original sequence is calculated in each simulation. ,in The original pulse value, For the reconstructed pulse value, if Exceeding a set threshold of 0.1 is considered a privacy breach. After 10,000 simulations, the number of breaches is counted and the privacy breach probability is calculated, assumed to be 0.18. The same simulation calculation is performed on member inference attacks and attribute inference attacks, yielding privacy breach probabilities of 0.12 and 0.15, respectively. Based on these probabilities, the privacy breach propagation path and impact range are analyzed. A privacy-efficiency optimization objective is constructed with privacy protection level (taking the negative of the weighted sum of the probabilities of each attack), communication bandwidth consumption (calculated based on the number of pulses and transmission time), and computational latency (calculated based on the system's pulse processing time) as targets. Finally, the NSGA-II algorithm is used to solve this optimization objective, obtaining a Pareto optimal solution set. From this set, a suitable scheme is selected to generate a privacy-efficiency protection optimization strategy for multi-agent pulse communication.

[0078] Furthermore, step S1 includes the following steps:

[0079] Step S11: Obtain the continuous input signal sequence corresponding to the multi-agent system;

[0080] Step S12: Generate input signal pulses from the continuous input signal sequence corresponding to the multi-agent system to generate a multi-agent input signal pulse sequence;

[0081] Step S13: The time-coded sparse matrix generation method performs time-delay sparse matrix transformation on the multi-agent input signal pulse sequence to generate a multi-agent signal sparse matrix with time-delay characteristics;

[0082] Step S14: The sparse matrix generation method based on rate coding performs pulse rate sparse matrix transformation on the multi-agent input signal pulse sequence to generate a multi-agent signal sparse matrix with rate characteristics.

[0083] Step S15: Perform spatiotemporal sparse pulse transformation on the multi-agent input signal pulse sequence based on the multi-agent signal sparse matrix with time delay and rate characteristics to generate a multi-agent spatiotemporal sparse pulse sequence.

[0084] As an embodiment of the present invention, reference Figure 2 As shown, Figure 1 A detailed flowchart of step S1 is shown below. In this embodiment, step S1 includes the following steps:

[0085] Step S11: Obtain the continuous input signal sequence corresponding to the multi-agent system;

[0086] In this embodiment of the invention, in a multi-agent system consisting of eight agents, the state data of each agent is collected in real time by a sensor array to obtain a continuous input signal sequence corresponding to the multi-agent system. Taking agent A1 as an example, its voltage sensor continuously collects its own operating voltage data at a sampling frequency of 1000Hz, obtaining 1000 consecutive voltage values ​​within 1 second. For example, the voltage values ​​at the first to fifth sampling points are 3.2V, 3.25V, 3.3V, 3.28V, and 3.32V, respectively, forming a continuous voltage signal sequence. The other seven agents also collect different types of state data, such as current and temperature, through their respective sensors at the same sampling frequency, which together constitute the continuous input signal sequence of the multi-agent system. These data are transmitted to the data processing center for subsequent processing via wired transmission lines using a fixed communication protocol.

[0087] Step S12: Generate input signal pulses from the continuous input signal sequence corresponding to the multi-agent system to generate a multi-agent input signal pulse sequence;

[0088] In this embodiment of the invention, input signal pulses are generated from the previously acquired continuous input signal sequence of the multi-agent system. The pulse generation threshold for voltage signals is set to 3.3V. Taking the voltage signal sequence of agent A1 as an example, when the voltage value of a sampling point first exceeds 3.3V, it is determined as the start point of a pulse. When the voltage value falls below 3.3V again, it is determined as the end point of the pulse. In the above 1-second signal sequence, the voltage value of the 3rd sampling point is 3.3V, and the voltage value of the 8th sampling point drops to 3.29V. Therefore, the 3rd to 8th sampling points constitute a pulse, which contains 6 sampling point data. Using the same threshold judgment method, pulse generation is performed on various continuous input signal sequences of all agents. For current signals, the threshold is set to 5A; for temperature signals, the threshold is set to 40℃. After processing, the current signal sequence of agent A2 generates 3 pulses, and the temperature signal sequence of agent A3 generates 1 pulse. Finally, a complete multi-agent input signal pulse sequence is generated. This sequence contains pulse information of different types of signals from each agent, providing basic data for subsequent analysis.

[0089] Step S13: The time-coded sparse matrix generation method performs time-delay sparse matrix transformation on the multi-agent input signal pulse sequence to generate a multi-agent signal sparse matrix with time-delay characteristics;

[0090] In this embodiment of the invention, a time-delay sparse matrix transformation is performed on the multi-agent input signal pulse sequence using a time-encoding-based sparse matrix generation method. Assuming the multi-agent system contains 8 agents, the time is divided into multiple intervals with a 100-millisecond time window. For a pulse from agent A1, if it occurs within the 200-300 millisecond time window, and a related pulse from agent A2 occurs within the 250-350 millisecond time window, the overlap duration of the two pulse time windows is calculated to be 50 milliseconds. An 8x8 matrix is ​​constructed, with rows and columns corresponding to the 8 agents. For pulses between any two agents, the overlap of their time windows is calculated as the matrix element value. If the overlap duration of the pulse time windows of agents A1 and A2 is 50 milliseconds, and the total time window length is 100 milliseconds, then the element value corresponding to row A1 and column A2 in the matrix is ​​50 / 100. =0.5 indicates the degree of temporal correlation between the two pulses. By traversing the pulse combinations between all agents in the multi-agent input signal pulse sequence, the matrix elements are calculated and filled, and finally a multi-agent signal sparse matrix with time delay characteristics is generated. This matrix reflects the time delay relationship of each agent's pulse in the time dimension.

[0091] Step S14: The sparse matrix generation method based on rate coding performs pulse rate sparse matrix transformation on the multi-agent input signal pulse sequence to generate a multi-agent signal sparse matrix with rate characteristics.

[0092] In this embodiment of the invention, a sparse matrix generation method based on rate coding is used to transform the pulse sequence of multi-agent input signals into a pulse rate sparse matrix. This allows for the counting of the number of pulses generated by each agent within a unit time (1 second), which is taken as the pulse rate of that agent. Agent A1 generated 5 pulses within 1 second, with a pulse rate of 5 pulses / second; Agent A2 generated 3 pulses, with a pulse rate of 3 pulses / second. An 8x8 matrix is ​​constructed, with rows and columns corresponding to the 8 agents. For the elements in the matrix, the formula is used... (in The first in the matrix Line 1 Column element values, For the first Pulse rate of each agent For the first Calculate the pulse rate of each agent. If we are calculating the matrix element values ​​corresponding to agents A1 and A2... =(5+3) / 4=4. By calculating the pulse rate combination among all agents, the matrix elements are filled to generate a multi-agent signal sparse matrix with rate characteristics. This matrix reflects the relationship between the pulse rates of each agent and provides a basis for subsequent signal processing.

[0093] Step S15: Perform spatiotemporal sparse pulse transformation on the multi-agent input signal pulse sequence based on the multi-agent signal sparse matrix with time delay and rate characteristics to generate a multi-agent spatiotemporal sparse pulse sequence.

[0094] In this embodiment of the invention, a spatiotemporal sparse pulse transformation is performed on the multi-agent input signal pulse sequence based on a previously generated multi-agent signal sparse matrix with time delay and rate characteristics. A time filtering threshold of 0.4 (corresponding to the matrix element value in step S13) and a rate filtering threshold of 3 (corresponding to the matrix element value in step S14) are set. For a pulse of agent A1, the element values ​​associated with its pulses in the time delay sparse matrix and the element values ​​corresponding to its pulses in the rate sparse matrix are examined. If the element value corresponding to A1 and A3 in the time delay sparse matrix is ​​0.3, which is less than the time delay threshold, the transformation is performed. The filtering threshold is 0.4; if the corresponding element value in the rate sparse matrix is ​​2.5, which is less than the rate filtering threshold of 3, then the pulse is determined to be a non-critical pulse and is removed. By performing the above filtering operation on all pulses in the multi-agent input signal pulse sequence, critical pulses that meet the threshold conditions are retained, and finally a multi-agent spatiotemporal sparse pulse sequence is generated. For example, if the original pulse sequence contains 50 pulses, after filtering, 20 critical pulses are retained. These pulses have more significant characteristics in the time and rate dimensions, effectively reducing signal redundancy and improving the privacy protection performance and transmission efficiency of multi-agent communication signals.

[0095] Furthermore, step S13 includes the following steps:

[0096] Step S131: Obtain the corresponding signal pulse time interval and signal pulse waveform characteristics by inputting the signal pulse sequence through multiple agents;

[0097] Step S132: Determine the dilution time window based on the signal pulse time interval and signal pulse waveform characteristics to obtain the size of the sparse signal pulse time window related to the original signal time distribution of the multi-agent agent;

[0098] Step S133: Based on the size of the sparse time window of the signal pulse related to the time distribution of the original signal of the multi-agent agent, perform pulse amplitude and time sequence position statistics for each sparse time window moment in the input signal pulse sequence of the multi-agent agent to obtain the multi-agent agent signal pulse amplitude and multi-agent agent signal time sequence position corresponding to each sparse time window moment.

[0099] Step S134: Calculate the time delay distribution difference at different times based on the amplitude of the multi-agent signal pulse and the timing position of the multi-agent signal corresponding to each sparse time window, and obtain the time delay distribution difference between the multi-agent signal pulses at different sparse time windows.

[0100] Step S135: Based on the time delay distribution differences between multi-agent signal pulses at different sparse time windows and combined with the time-coded sparse matrix generation method, the multi-agent input signal pulse sequence is transformed into a time delay sparse matrix to generate a multi-agent signal sparse matrix with time delay characteristics.

[0101] As an embodiment of the present invention, reference Figure 3 As shown, Figure 2 A detailed flowchart of step S13 is shown in this embodiment. Step S13 includes the following steps:

[0102] Step S131: Obtain the corresponding signal pulse time interval and signal pulse waveform characteristics by inputting the signal pulse sequence through multiple agents;

[0103] In this embodiment of the invention, taking a network composed of 10 agents as an example, the input signal pulse sequence of multiple agents is obtained. Assuming that agent A1 emits a signal pulse sequence within 1 second, pulses are generated at 0.1 seconds, 0.3 seconds, 0.6 seconds, and 0.8 seconds respectively. A high-precision timestamp recording device is used to accurately record the time of each pulse's generation, thereby calculating the signal pulse time interval. For example, the time interval between the first and second pulses is 0.3 − 0.1 = 0.2 seconds, and the time interval between the second and third pulses is 0.6 − 0.3 = 0.3 seconds. Simultaneously, waveform acquisition devices such as oscilloscopes are used to acquire the signal pulse waveform characteristics. Taking voltage signals as an example, parameters such as the peak voltage, rise time, and fall time of each pulse are recorded. For example, the first pulse of agent A1 has a peak voltage of 5 volts, a rise time of 0.01 seconds, and a fall time of 0.02 seconds. The corresponding signal pulse time intervals and signal pulse waveform characteristics are obtained for the input signal pulse sequences of all 10 agents in the network in this manner, forming a complete original signal feature dataset.

[0104] Step S132: Determine the dilution time window based on the signal pulse time interval and signal pulse waveform characteristics to obtain the size of the sparse signal pulse time window related to the original signal time distribution of the multi-agent agent;

[0105] In this embodiment of the invention, a dilution time window is determined based on previously acquired signal pulse time intervals and signal pulse waveform characteristics. Then, a statistical analysis method is used to calculate the average value of the signal pulse time intervals for all agents. and standard deviation Assuming the average time interval of the signal pulses from 10 agents is calculated =0.25 seconds, standard deviation =0.05 seconds, the formula for calculating the size of the signal pulse sparse time window is as follows: (in The adjustment coefficient is set according to the required signal sparsity, which is also the case here. The size of the sparse time window of the signal pulse is then determined. =2×0.25=0.5 seconds. The size of this time window is related to the time distribution of the original signal of the multi-agent system. In this way, the continuous signal pulse sequence is divided into multiple time windows for subsequent pulse information statistics and analysis. This can ensure that key signal features are captured while reducing the amount of data processing.

[0106] Step S133: Based on the size of the sparse time window of the signal pulse related to the time distribution of the original signal of the multi-agent agent, perform pulse amplitude and time sequence position statistics for each sparse time window moment in the input signal pulse sequence of the multi-agent agent to obtain the multi-agent agent signal pulse amplitude and multi-agent agent signal time sequence position corresponding to each sparse time window moment.

[0107] In this embodiment of the invention, the input signal pulse sequence of multiple agents is processed based on a previously determined 0.5-second sparse time window size. Taking the signal pulse sequence of agent A1 as an example, within the first sparse time window of 0-0.5 seconds, pulses are generated at 0.1 seconds and 0.3 seconds. The amplitudes of these two pulses are counted, assuming they are 5 volts and 4.8 volts respectively. At the same time, their temporal positions within this time window are recorded, i.e., the positions of 0.1 seconds and 0.3 seconds relative to time 0. For the signal pulse sequences of all 10 agents, the pulse amplitude and temporal position are counted at each 0.5-second sparse time window. For example, within the 0.5-1.0 second time window, agent A2 generates a pulse at 0.6 seconds with an amplitude of 5.2 volts. Its amplitude is recorded as 5.2 volts, and its temporal position is 0.1 seconds (relative to time 0.5 seconds). Finally, the multi-agent signal pulse amplitude and multi-agent signal temporal position corresponding to each sparse time window are obtained, forming a detailed pulse information statistics table.

[0108] Step S134: Calculate the time delay distribution difference at different times based on the amplitude of the multi-agent signal pulse and the timing position of the multi-agent signal corresponding to each sparse time window, and obtain the time delay distribution difference between the multi-agent signal pulses at different sparse time windows.

[0109] In this embodiment of the invention, the time delay distribution difference at different times is calculated based on the amplitude of the multi-agent signal pulses and the timing position of the multi-agent signals corresponding to each sparse time window obtained previously. Taking two adjacent sparse time windows (0-0.5 seconds and 0.6-1.0 seconds) as an example, assuming that agent A1 generates a pulse at 0.1 seconds in the first time window and at 0.6 seconds in the second time window, the time delay between these two pulses is calculated as follows: Seconds, using the variance formula (in For the time delay between different pulses, For average delay, The time delay distribution difference is calculated using the number of pulse pairs. Assuming that within two time windows, agents A1, A2, and A3 each have two pairs of pulses, the calculated time delays are 0.5 seconds, 0.4 seconds, and 0.6 seconds respectively. The average time delay is... =0.5 + 0.4 + 0.6 / 3 = 0.5 seconds, then the time delay distribution difference =0+0.01+0.01 / 3≈0.0067. For all agents, the time delay distribution difference is calculated according to this method for pulses at different sparse time windows, and finally comprehensive time delay distribution difference data is obtained.

[0110] Step S135: Based on the time delay distribution differences between multi-agent signal pulses at different sparse time windows and combined with the time-coded sparse matrix generation method, the multi-agent input signal pulse sequence is transformed into a time delay sparse matrix to generate a multi-agent signal sparse matrix with time delay characteristics.

[0111] In this embodiment of the invention, the input signal pulse sequence of multiple agents is transformed by combining the time delay distribution differences between signal pulses of multiple agents at different sparse time windows obtained previously, and a time-coded sparse matrix generation method. The rows of the sparse matrix represent different agents (10 rows, corresponding to 10 agents), and the columns represent different sparse time windows (assuming 20 time windows, corresponding to 20 columns). For each agent at each time window, if a pulse exists, the time delay distribution difference value between that pulse and other related pulses is filled into the corresponding position in the matrix; if no pulse exists, 0 is filled in. If agent A1 has a pulse in the first time window and the time delay distribution difference between it and subsequent related pulses is 0.0067, then 0.0067 is filled in the first row and first column of the sparse matrix; if there is no pulse in the second time window, then 0 is filled in the first row and second column. In this way, the entire multi-agent input signal pulse sequence is transformed into a 10-row, 20-column multi-agent signal sparse matrix with time delay characteristics. This matrix retains the key time delay information of the signal pulses and is stored in a sparse form, which facilitates subsequent multi-agent privacy protection analysis based on the matrix, such as mining privacy leakage risk points in the signal through matrix operations.

[0112] Furthermore, step S14 includes the following steps:

[0113] Step S141: Perform statistical calculations on the pulse frequency and amplitude within each signal pulse window of the multi-agent input signal pulse sequence to obtain the corresponding multi-agent signal pulse frequency and multi-agent signal pulse amplitude within each signal pulse window;

[0114] In this embodiment of the invention, taking a network of 10 agents as an example, the input signal pulse sequence of multiple agents is processed. The duration of each signal pulse window is set to 0.2 seconds. Taking the signal pulse sequence of agent A1 (which generates pulses at 0.1 seconds, 0.3 seconds, 0.6 seconds, and 0.8 seconds respectively) as an example, in the first signal pulse window of 0-0.2 seconds, there is one pulse. According to the pulse frequency calculation formula... (in The number of pulses within the window. (where the window duration is used), the frequency of multi-agent signal pulses within this window can be obtained. =5Hz; the pulse amplitude is 5 volts, meaning the multi-agent signal pulse amplitude within this window is 5 volts. For the signal pulse sequence of all 10 agents, the signal pulse window is divided sequentially at 0.2-second intervals. For example, if agent A2 has 0 pulses within the 0.2-0.4 second signal pulse window, its pulse frequency is... =0Hz, pulse amplitude is recorded as 0. In this way, the number of pulses and amplitude of each agent in each signal pulse window are statistically calculated, and finally a complete dataset of the multi-agent signal pulse frequency and multi-agent signal pulse amplitude in each signal pulse window is obtained, providing basic data for subsequent analysis.

[0115] Step S142: Obtain the corresponding multi-agent pulse signal intensity through the multi-agent input signal pulse sequence, and perform pulse rate modulation on the multi-agent input signal pulse sequence based on the multi-agent pulse signal intensity to generate the modulation signal pulse rate corresponding to the multi-agent signal pulse.

[0116] In this embodiment of the invention, the pulse signal intensity of multiple agents is obtained through the input signal pulse sequence of multiple agents. For agent A1, assuming its signal pulse amplitudes are 5 volts, 4.8 volts, 5.2 volts, and 5 volts, the formula is used. (in For signal strength, The number of pulses. For the first Calculate the pulse signal strength using the pulse amplitude values. 5 + 4.8 + 5.2 + 5 / 4 = 5 volts. Similarly, calculate the pulse signal strength of the other 9 agents. Based on the multi-agent pulse signal strength, perform pulse rate modulation on the multi-agent input signal pulse sequence, setting the modulation formula as follows: (in The modulated pulse rate, Let be the modulation coefficient, set here. =2), taking agent A1 as an example, its modulated pulse rate =2×5=10 times / second, which is the modulation signal pulse rate corresponding to the multi-agent signal pulse. The signal pulse sequences of all 10 agents in the network are modulated in this way to obtain the modulation signal pulse rate corresponding to each agent. This is used to adjust the transmission rate of the signal pulse and prepare for the construction of a matrix with rate characteristics.

[0117] Step S143: Based on the modulation signal pulse rate corresponding to the multi-agent signal pulse and combined with the sparse matrix generation method of rate coding, the pulse rate sparse weighted reconstruction of the multi-agent signal pulse frequency and the multi-agent signal pulse amplitude corresponding to each signal pulse window is performed to generate a multi-agent signal sparse matrix with rate characteristics.

[0118] In this embodiment of the invention, based on the previously obtained modulation signal pulse rate corresponding to the multi-agent signal pulses, and combined with a rate-encoded sparse matrix generation method, pulse rate sparse weighted reconstruction is performed on the frequency and amplitude of the multi-agent signal pulses corresponding to each obtained signal pulse window. The rows of the sparse matrix represent different agents (10 rows in total, corresponding to 10 agents), and the columns represent different signal pulse window times (assuming 50 window times in total, corresponding to 50 columns). For each agent at each signal pulse window time, a weighting formula is used. (in The first in the matrix Line 1 Column element values, , Here are the weighting coefficients. =0.4, =0.6, For the first The agent in the th... The pulse frequency of each window, For the first The agent in the th... The pulse amplitude of each window, For the first Modulation signal pulse rate of an agent (The maximum value among all agent modulation signal pulse rates), assuming the modulation signal pulse rate of agent A1 is... =10 times / second, within the first signal pulse window, the pulse frequency is... =5Hz, pulse amplitude =5 volts, the maximum modulation signal pulse rate among all intelligent agents =12 times / second, then the corresponding element in the matrix =0.4×5+0.6×5×10 / 12=2+2.5=4.5. If an agent has no pulse in a certain window, the corresponding matrix element value is 0. In this way, the pulse frequency, amplitude and modulated pulse rate of each agent in each signal pulse window are weighted and calculated, and filled into the sparse matrix to generate a 10-row, 50-column multi-agent signal sparse matrix with rate characteristics. This matrix integrates the frequency, amplitude and rate information of multi-agent signals and stores them in a sparse form, which is convenient for subsequent multi-agent privacy protection analysis based on the matrix, such as mining potential privacy leakage patterns through matrix feature extraction.

[0119] Furthermore, step S15 includes the following steps:

[0120] Step S151: Based on the sparse matrix of multi-agent signals with time delay characteristics, perform time delay sparsification filtering on the pulse sequence of multi-agent input signals to obtain the multi-agent time delay sparse filtered pulse sequence.

[0121] In this embodiment of the invention, a time delay sparsity filter is applied to the multi-agent input signal pulse sequence based on a pre-generated multi-agent signal sparse matrix with time delay characteristics. Assuming the sparse matrix is ​​10 rows and 20 columns, with each row corresponding to 10 agents and each column corresponding to 20 time windows, the matrix element values ​​represent the time delay distribution differences of the signal pulses from different agents within their respective time windows. A filtering threshold of 0.01 is set. For each pulse in the multi-agent input signal pulse sequence, the corresponding element value in the sparse matrix is ​​searched based on the agent and time window it belongs to. Taking the pulse of agent A1 in the 3rd time window as an example, if the element value at the corresponding position (1st row, 3rd column) in the sparse matrix is ​​0... If the value of a pulse is less than 0.01, it is retained. If the value of a matrix element corresponding to an agent in a certain time window is greater than or equal to 0.01, such as agent A2 having a value of 0.012 in the 5th time window, the pulse is discarded. By traversing the pulses of all agents in each time window in the multi-agent input signal pulse sequence and filtering according to the above rules, the multi-agent time delay sparse filtered pulse sequence is finally obtained. For example, if there are 100 pulses in the original sequence from agents A1 to A10, after filtering, 70 pulses are retained to form a new pulse sequence. This sequence is sparsified in the time dimension, reducing redundant pulses and highlighting key pulses with obvious time delay characteristics.

[0122] Step S152: Based on the sparse matrix of the multi-agent signal with rate characteristics, the multi-agent time delay sparse filter pulse sequence is rate-sparsely modulated in spatial location to generate a multi-agent rate sparse modulation pulse sequence.

[0123] In this embodiment of the invention, a rate-sparse modulation is performed on the previously obtained multi-agent time-delay sparse filter pulse sequence based on a rate-featured multi-agent signal sparse matrix. This rate-feature sparse matrix is ​​10 rows and 50 columns, where each row represents 10 agents and each column represents 50 signal pulse window times. The matrix element values ​​are calculated by weighting the pulse frequency, amplitude, and modulated pulse rate. For each pulse in the multi-agent time-delay sparse filter pulse sequence, the element value at the corresponding position in the sparse matrix is ​​obtained according to its agent and window time. Taking the pulse of agent A3 at the 10th window time as an example, assuming this pulse is within the rate-feature sparse matrix... The corresponding element value in the array is 3.8. The modulation rule is set as follows: when the element value is greater than 3, the spatial transmission path of the pulse is adjusted to a higher priority path; when the element value is less than or equal to 3, the original transmission path is maintained. Therefore, the transmission path of the pulse is adjusted to a higher priority line for transmission. For all pulses in the multi-agent time delay sparse filtering pulse sequence, the spatial transmission path is adjusted according to the element value of the rate feature sparse matrix in accordance with this rule. Finally, a multi-agent rate sparse modulation pulse sequence is generated. In this way, the pulse is modulated according to the rate characteristics of the signal in terms of spatial location distribution, which optimizes the pulse transmission path and improves the efficiency and relevance of signal transmission.

[0124] Step S153: Obtain the pulse energy magnitude corresponding to each pulse signal through the multi-agent rate sparse modulation pulse sequence, and evaluate the pulse attenuation based on the pulse energy magnitude corresponding to each pulse signal to obtain the pulse attenuation factor corresponding to the multi-agent pulse signal.

[0125] In this embodiment of the invention, the pulse energy magnitude corresponding to each pulse signal is obtained through a multi-agent rate sparse modulation pulse sequence. For a certain pulse of agent A1, assuming its amplitude is 4.5 volts and its duration is 0.01 seconds, the pulse energy is calculated according to the pulse energy calculation formula. (Assuming the pulse signal can be equivalent to a capacitor discharge model, the capacitor...) (Faraday), the pulse energy can be obtained. Joule calculates the energy of each pulse from all agents in a multi-agent rate-sparsely modulated pulse sequence according to this formula, and evaluates pulse attenuation based on the pulse energy of each pulse signal. The attenuation evaluation formula is set as follows: (in The pulse attenuation factor. (assuming the minimum value among all pulse energies) Joule, for the pulse of the aforementioned agent A1, its pulse attenuation factor By calculating the attenuation factor for all pulses, the final set of pulse attenuation factors corresponding to the multi-agent pulse signal is obtained. This factor reflects the degree of attenuation of each pulse relative to the minimum energy pulse.

[0126] Step S154: Based on the pulse attenuation factor corresponding to the multi-agent pulse signal, optimize the spatiotemporal pulse distribution of each pulse signal in the multi-agent rate sparse modulation pulse sequence to generate a multi-agent spatiotemporal sparse pulse sequence.

[0127] In this embodiment of the invention, the spatiotemporal pulse distribution of each pulse signal in the multi-agent rate sparse modulation pulse sequence is optimized based on the pulse attenuation factor corresponding to the multi-agent pulse signal. The optimization rules are as follows: when the pulse attenuation factor is less than 0.8, the pulse is delayed by 0.02 seconds in time and adjusted to a backup transmission path in space; when the pulse attenuation factor is greater than or equal to 0.8, the spatiotemporal position of the pulse remains unchanged. Taking a pulse from agent A4 as an example, its pulse attenuation factor is 0.75. According to the rules, the pulse is delayed from the original 0.5 second to 0.52 seconds in time. At a given second, the transmission path is adjusted spatially from the original path to the backup path. For pulses with an attenuation factor of 0.85 for agent A5, their spatiotemporal position is maintained. All pulses in the multi-agent rate sparse modulation pulse sequence are adjusted in time and space according to their respective pulse attenuation factors, following this rule, ultimately generating a multi-agent spatiotemporal sparse pulse sequence. This sequence optimizes the distribution of pulses in both time and space, reduces interference from weaker pulses, enhances the stability of signal transmission and privacy protection, and makes multi-agent communication signals more rational and efficient in the spatiotemporal dimension.

[0128] Furthermore, step S2 includes the following steps:

[0129] Step S21: Construct a corresponding hierarchical spatiotemporal spiking neural network by adopting a structure that combines recurrent neural networks and long short-term memory networks. Specifically, it includes a bottom layer, a middle layer, and a top layer. The bottom layer is used to process the original spiking sequence, the middle layer is used to extract the corresponding spiking temporal features, and the top layer is used to make decisions and output the corresponding communication spiking sequence.

[0130] In this embodiment of the invention, a hierarchical spatiotemporal spiking neural network is constructed, employing a structure combining recurrent neural networks and long short-term memory networks. The bottom layer consists of 10 neurons, used to process the original spiking sequence. Each neuron corresponds to an agent, receiving a multi-agent spatiotemporally sparse spiking sequence as input. The middle layer contains 20 neurons, used to extract spiking temporal features. Each neuron is connected to all neurons in the bottom layer. The top layer has 5 neurons, used to make decisions and output communication spiking sequences, receiving the output of the middle layer and generating the final result. The bottom layer neurons adopt the Integrate-and-Fire (IF) model, and its membrane potential update formula is: ,in =0.8 is the membrane potential time constant. For connection weights, For input pulse, As the threshold, when a middle layer neuron receives a pulse from the lower layer, the temporal information is processed by the LSTM unit. The calculation formulas for its input gate, forget gate, and output gate are as follows: , , ,in For the sigmoid function, This is the weight matrix. As a bias term, the top-level neuron uses the softmax function to make a decision output based on the temporal features extracted from the middle layer, with the formula as follows: ,in For the first The input of each neuron, This is the number of output categories.

[0131] Step S22: Input the multi-agent spatiotemporal sparse pulse sequence into the bottom layer of the hierarchical spatiotemporal spiking neural network for pulse preprocessing, and input the preprocessed multi-agent spatiotemporal sparse pulse sequence into the middle layer of the hierarchical spatiotemporal spiking neural network to train the pulse extraction network by introducing a pulse temporal attention mechanism. At the same time, during the training process, a pulse network training method based on differential privacy is designed and constructed to introduce Gaussian noise into the spiking neural network and prevent gradient inversion attacks by combining the gradient pruning strategy in the training process, so as to train and generate the pulse temporal feature output corresponding to each agent.

[0132] In this embodiment of the invention, a multi-agent spatiotemporal sparse pulse sequence is input into the bottom layer of a hierarchical spatiotemporal spiking neural network for preprocessing. The bottom layer neurons integrate the input pulses, and when the membrane potential exceeds a threshold, an output pulse is generated. The preprocessed pulse sequence is then passed to the middle layer, where a pulse temporal attention mechanism is introduced to train the pulse extraction network. The attention weight calculation formula is as follows: ,in , The current hidden state. For the first The characteristic vector of each pulse , , To train the weight matrix, a differential privacy-based spiking network training method is designed during training. Gaussian noise is added to the gradient at each parameter update, and the standard deviation of the noise is... ,in Sensitivity to gradients, = 0.5 is the privacy budget parameter, and a gradient pruning strategy is used to prune the gradient norm to 0.5. =1.0, to prevent gradient inversion attacks, the training objective function is the cross-entropy loss function: ,in For real labels, To predict probabilities, the network parameters are updated using the stochastic gradient descent algorithm, with the learning rate set to [value missing]. = 0.01, after 100 training cycles, the pulse timing feature output corresponding to each agent is finally generated.

[0133] Step S23: Calculate the pulse interaction frequency and pulse sensitivity between agents by using the pulse timing features output from the top layer of the hierarchical spatiotemporal spiking neural network. Then, dynamically configure the corresponding privacy budget based on the pulse interaction frequency and pulse sensitivity between agents. Parameters, while based on this privacy budget The parameters are used to make decisions at the top level, resulting in the corresponding communication pulse sequence, which generates the multi-agent time-series key communication pulse sequence.

[0134] In this embodiment of the invention, the pulse interaction frequency and pulse sensitivity between agents are calculated at the top layer of the hierarchical spatiotemporal spiking neural network based on the pulse timing characteristics output corresponding to each agent. The formula for calculating the pulse interaction frequency is as follows: ,in and respectively intelligent agents and At any moment pulse output, The formula for calculating pulse sensitivity is: (Total time steps) ,in The loss function is defined by dynamically configuring the privacy budget parameter based on the impulse interaction frequency and impulse sensitivity. For agent pairs with high interaction frequency and high sensitivity, a smaller privacy budget parameter is assigned. The calculation formula is as follows: ,in =1.0 is the initial privacy budget. When the top-level decision outputs the communication pulse sequence, noise is injected into the output according to the privacy budget parameter to generate the multi-agent time-series key communication pulse sequence. The noise injection formula is: ,in This is the original output. , To improve the sensitivity of the output, this method effectively protects the privacy of intelligent agents while ensuring communication quality.

[0135] Furthermore, step S3 includes the following steps:

[0136] Step S31: Perform pulse timing perturbation analysis on the multi-agent time-series key communication pulse sequences to obtain the distribution characteristics of multi-agent communication pulse timing perturbation;

[0137] In this embodiment of the invention, pulse timing perturbation analysis is performed on the timing-critical communication pulse sequences of a multi-agent system. Assuming there are five agents (A, B, C, D, and E) in the multi-agent system, taking a pulse sequence of agent A as an example, the pulse generation times are 0.1 seconds, 0.3 seconds, 0.5 seconds, 0.7 seconds, and 0.9 seconds, respectively. By calculating the time intervals between adjacent pulses, the time interval sequence is obtained as 0.2 seconds, 0.2 seconds, 0.2 seconds, and 0.2 seconds. Statistical analysis methods are then used to calculate the average value of these time intervals. and standard deviation ,average value =0.2 + 0.2 + 0.2 + 0.2 / 4 = 0.2 seconds, standard deviation =0 seconds, indicating that the pulse timing of agent A is relatively stable with small perturbations. The same analysis was performed on the pulse sequences of the other four agents. The pulse time interval sequences of agent B were 0.15 seconds, 0.25 seconds, 0.18 seconds, and 0.22 seconds, and the average value was calculated. = 0.2 seconds, standard deviation The value is approximately 0.035 seconds, indicating that the pulse timing of agent B has a certain disturbance. By analyzing the pulse sequences of all agents, the distribution characteristics of pulse timing disturbance in multi-agent communication are obtained. That is, the degree of pulse timing disturbance varies among different agents. The pulse timing of some agents is relatively stable, while the pulse timing of others fluctuates greatly.

[0138] Step S32: Obtain the pulse communication disturbance delay and pulse communication disturbance frequency difference between each agent through the pulse timing disturbance distribution characteristics of multi-agent communication, and perform pulse time jitter amplitude statistics on each pulse signal in the multi-agent timing key communication pulse sequence based on the pulse communication disturbance delay and pulse communication disturbance frequency difference between each agent to obtain the communication pulse signal time jitter amplitude between each agent.

[0139] In this embodiment of the invention, based on the previously obtained multi-agent communication pulse timing perturbation distribution characteristics, the pulse communication perturbation delay and pulse communication perturbation frequency difference between each agent are obtained. Taking agents A and B as examples, assuming that a pulse from agent A is generated at 0.3 seconds and a related pulse from agent B is generated at 0.35 seconds, the pulse communication perturbation delay is 0.35 - 0.3 = 0.05 seconds. The pulse communication perturbation frequency difference is calculated, with a 1-second time window, and the number of pulses from agents A and B within this time window is counted. Assuming that agent A generates 5 pulses in 1 second and agent B generates 4 pulses in 1 second, the pulse communication perturbation frequency difference is |5 - 4| = 1 pulse / second. Based on the pulse communication perturbation delay and pulse communication perturbation frequency difference between each agent, the pulse time jitter amplitude of each pulse signal in the multi-agent timing key communication pulse sequence is statistically analyzed, and the jitter amplitude calculation formula is set as follows: ,in For pulse communication disturbance delay, To address the frequency difference in pulse communication perturbations, let's take a pulse pair between agents A and B as an example. =0.05 seconds, =1 time / second, then the pulse time jitter amplitude =1.05. By calculating the pulse pairs between all agents, the time jitter amplitude of the communication pulse signals between each agent is obtained. For example, the pulse time jitter amplitude between agents A and C is 0.8, and the pulse time jitter amplitude between agents B and D is 1.2, etc.

[0140] Step S33: Generate a corresponding multi-agent pulse time jitter amplitude map based on the time jitter amplitude of the communication pulse signals between each agent, and perform pulse selective deletion optimization on each pulse signal in the multi-agent time-series key communication pulse sequence based on the multi-agent pulse time jitter amplitude map, so as to filter and delete redundant pulse signals with consistent time jitter amplitude of communication pulse signals, so as to generate a multi-agent pulse optimization deletion sequence.

[0141] In this embodiment of the invention, a multi-agent pulse time jitter amplitude map is generated based on the previously obtained communication pulse signal time jitter amplitude between the various agents. A bar chart is drawn with the agents as the horizontal axis and the pulse time jitter amplitude as the vertical axis. For example, the pulse time jitter amplitude of agent A is 0.5, that of agent B is 0.8, that of agent C is 0.6, that of agent D is 0.9, and that of agent E is 0.7. Pulse signals with the same pulse time jitter amplitude are marked on the map. Assuming that the pulse time jitter amplitude of agents A and C is 0.6 in a certain time period, these pulse signals are redundant pulse signals. Based on the multi-agent pulse time jitter amplitude map, pulse selective deletion optimization is performed on each pulse signal in the multi-agent time-critical communication pulse sequence. Redundant pulse signals with the same communication pulse signal time jitter amplitude are filtered out and deleted. By traversing the pulse sequence, the time jitter amplitude of each pulse is compared with the data in the map, and redundant pulses are deleted. For example, in a sequence containing 100 pulses, after comparison and filtering, 20 redundant pulses are removed, and a multi-agent pulse optimization deletion sequence is finally generated. This sequence reduces redundant information and improves the effectiveness and privacy protection performance of the pulse sequence.

[0142] Step S34: Perform pulse perturbation distribution randomization processing on the timing of the corresponding remaining pulse signals in the multi-agent pulse optimization deletion sequence to generate a multi-agent pulse perturbation randomization sequence.

[0143] In this embodiment of the invention, the timing of the remaining pulse signals corresponding to the previously generated multi-agent pulse optimization deletion sequence is randomized by pulse perturbation distribution. The randomization range is set to ±0.1 seconds. Taking a remaining pulse of agent A as an example, its original generation time is 0.5 seconds. A random number is generated within the randomization range, assuming it is 0.05 seconds. Then the new generation time of the pulse is 0.5 + 0.05 = 0.55 seconds. All remaining pulse signals in the multi-agent pulse optimization deletion sequence are randomized in this way. For example, a remaining pulse of agent B... The original pulse time is 0.7 seconds, and the random number is -0.08 seconds. Therefore, the new time is 0.7 - 0.08 = 0.62 seconds. By randomizing the time sequence of all remaining pulses, a multi-agent pulse perturbation randomized sequence is generated. This sequence has a more random temporal distribution, which increases the unpredictability of the pulse sequence and further improves the privacy protection performance of multi-agent communication. For example, the temporal distribution of pulses in the original optimized deletion sequence is relatively concentrated. After randomization, the temporal distribution of pulses is more dispersed, reducing the risk of obtaining privacy information by analyzing the pulse time sequence.

[0144] Furthermore, step S4 includes the following steps:

[0145] Step S41: Construct privacy leakage attack scenario models corresponding to pulse sequence reconstruction attack, membership reasoning attack, and attribute reasoning attack;

[0146] In this embodiment of the invention, when constructing a pulse sequence reconstruction attack scenario model, the attacker is assumed to be able to obtain a multi-agent pulse perturbation randomized sequence. Using known pulse generation rules and communication protocols, the attacker attempts to reconstruct the original pulse sequence using a maximum likelihood estimation-based reconstruction algorithm. Taking a system with four agents (A, B, C, D) as an example, assuming the original pulse sequence uses binary encoding, the attacker calculates the probability of generating the current perturbation sequence under different original sequence assumptions based on the received perturbation sequence, and selects the original sequence with the highest probability as the reconstruction result. In the member inference attack scenario model, the attacker uses historical pulse sequence data of the multi-agent system and the currently observed pulse perturbation randomized sequence to determine whether a certain agent belongs to the system using a similarity-based calculation method, for example, using the cosine similarity formula. Calculate the target agent's pulse sequence With the average pulse sequence of the intelligent agents in the system The similarity is set to a similarity threshold of 0.7. If the similarity exceeds this threshold, it is inferred that the target agent belongs to the system. In the attribute reasoning attack scenario model, the attacker constructs a decision tree-based classification model based on the statistical characteristics of the agent's pulse sequence (such as pulse frequency, pulse interval distribution, etc.) to infer the agent's attributes. Assuming that the agent has two attributes, "working mode 1" and "working mode 2", the attacker trains the decision tree model by analyzing a large amount of agent pulse sequence data with known attributes. When a new pulse perturbation randomized sequence is obtained, it is input into the model to predict the agent's attributes.

[0147] Step S42: Use Monte Carlo simulation combined with the corresponding privacy leakage attack scenario model to simulate privacy leakage attacks on the randomized sequence of multi-agent pulse perturbation, so as to generate the corresponding multi-agent pulse privacy leakage probability under different attack scenarios.

[0148] In this embodiment of the invention, a privacy leakage attack simulation is performed on a multi-agent pulse perturbation randomized sequence using Monte Carlo simulation combined with a pre-constructed privacy leakage attack scenario model. The number of Monte Carlo simulations is set to 10,000, with each attack scenario model simulated independently. In the pulse sequence reconstruction attack simulation, during each simulation, the attacker reconstructs the original pulse sequence using a maximum likelihood estimation reconstruction algorithm based on the current multi-agent pulse perturbation randomized sequence. Taking the pulse sequence of agent A as an example, if the original sequence is [1, 0, 1, 0, 1], the reconstructed sequence is [1, 1, 1, 0, 0]. The original sequence and the reconstructed sequence are compared, and the number of pulses with reconstruction errors is counted. After 10,000 simulations, the proportion of the number of reconstructed error pulses to the total number of pulses is calculated as the privacy leakage probability under this attack scenario. Assuming the calculation results... For member reasoning attack simulations, a random agent is selected as the target in each simulation. The cosine similarity algorithm is used to determine whether the agent belongs to the system. If the agent actually belongs to the system but is misjudged as not belonging, or if the agent actually does not belong but is misjudged as belonging, it is counted as a privacy leak. After 10,000 simulations, the number of privacy leaks is counted and divided by the total number of simulations to obtain the privacy leak probability in the member reasoning attack scenario, which is assumed to be 0.1. In attribute reasoning attack simulations, each simulation inputs a randomized sequence of impulse perturbation into a trained decision tree classification model for attribute prediction. If the prediction is incorrect, it is considered a privacy leak. After 10,000 simulations, the proportion of the number of prediction errors to the total number of simulations is calculated. It is assumed that the privacy leak probability in the attribute reasoning attack scenario is 0.12. Thus, the multi-agent impulse privacy leak probabilities corresponding to different attack scenarios are generated.

[0149] Step S43: Based on the multi-agent pulse privacy leakage probability under different attack scenarios, perform privacy leakage diffusion impact analysis on the multi-agent pulse perturbation randomization sequence to obtain the diffusion path and impact range of the corresponding pulse privacy leakage event in the multi-agent system;

[0150] In this embodiment of the invention, the privacy leakage propagation impact analysis of multi-agent pulse perturbation randomized sequences is performed based on the privacy leakage probability of multi-agent pulses under different attack scenarios. Taking the privacy leakage caused by pulse sequence reconstruction attack as an example, assuming that the original pulse sequence of agent A contains critical task instruction information, when the sequence is reconstructed incorrectly, due to the data interaction dependency relationship between agent A and agents B and C, the incorrect instruction information will be transmitted to agents B and C, causing deviations in their task execution, which in turn affects agent D, which has data interaction with B and C, forming a diffusion path of A→B→C→D. This is achieved by analyzing the data interaction topology and dependency relationship between the agents in the system. Based on the privacy leakage probability, the degree of impact on each agent is determined. The formula for calculating the degree of impact is set as I=p×d, where p is the privacy leakage probability under the corresponding attack scenario, and d is the agent's dependence in the data interaction topology (obtained by calculating the sum of the edge weights of the agent and other affected agents). Assuming that the dependence of agent B is d=3, and p=0.15 under the pulse sequence reconstruction attack scenario, its degree of impact I=0.15×3=0.45. The threshold for the scope of impact is set to 0.3, and agents with an impact degree greater than or equal to this threshold are included in the scope of impact. Finally, the diffusion path and scope of the corresponding pulse privacy leakage event in the multi-agent system are obtained.

[0151] Step S44: Based on the propagation path and impact range of the corresponding impulse privacy leakage event in the multi-agent system, construct a privacy-efficiency optimization objective corresponding to the privacy protection level, communication bandwidth consumption, and computational latency;

[0152] In this embodiment of the invention, a privacy-efficiency optimization objective is constructed based on the propagation path and impact range of a pulse privacy breach event. The level of privacy protection is measured by the negative of the weighted sum of the probabilities of privacy breaches under different attack scenarios. Let the weights of pulse sequence reconstruction attacks, membership inference attacks, and attribute inference attacks be respectively... =0.4、 =0.3、 =0.3, corresponding to a privacy leak probability of , , The level of privacy protection Communication bandwidth consumption is determined by calculating the bandwidth resources occupied by the multi-agent pulse perturbation randomized sequence during transmission, assuming that each pulse signal occupies... Bit bandwidth, number of pulses in the sequence Transmission time is Then the communication bandwidth consumption The computational delay is measured based on the time required for a multi-agent system to process a pulse signal, assuming the average time taken by the system to process a single pulse signal. Then calculate the delay. = Therefore, a privacy-efficiency optimization objective is constructed based on the level of privacy protection, communication bandwidth consumption, and computational latency, aiming to balance these three indicators and improve the performance of multi-agent systems.

[0153] Step S45: Based on the privacy-efficiency optimization objective corresponding to the privacy protection level, communication bandwidth consumption and computation delay, perform privacy leakage protection analysis on the randomized sequence of multi-agent pulse perturbation. Use the NSGA-II algorithm to solve for the Pareto optimal solution set corresponding to the privacy-efficiency optimization objective. Based on the Pareto optimal solution set, obtain the balance strategy between privacy and efficiency for pulse signal communication to generate the privacy-efficiency protection optimization strategy for multi-agent pulse communication.

[0154] In this embodiment of the invention, based on the privacy-efficiency optimization objective, the NSGA-II algorithm is used to solve for the corresponding Pareto optimal solution set. The parameters of the multi-agent impulse perturbation randomization sequence (such as the threshold in the impulse generation rule, the range of randomization processing, etc.) are used as decision variables and input into the NSGA-II algorithm. The algorithm generates multiple sets of impulse perturbation randomization sequence schemes with different parameter combinations through genetic operations such as selection, crossover, and mutation. For each scheme, its corresponding privacy protection level, communication bandwidth consumption, and computational delay are calculated. Non-dominated solutions are selected using non-dominated sorting and congestion calculation, and the Pareto optimal solution set is gradually constructed. Assuming that... After multiple rounds of iterative calculations, a Pareto optimal solution set containing 10 schemes is obtained. A suitable scheme is selected from this set as the balance strategy between privacy and efficiency for pulse signal communication. For example, Scheme 1 has a privacy protection level of -0.12, a communication bandwidth consumption of 10Mbps, and a computational delay of 50ms; Scheme 2 has a privacy protection level of -0.1, a communication bandwidth consumption of 12Mbps, and a computational delay of 45ms. Depending on the actual application requirements, if privacy protection is more important, Scheme 1 is chosen; if communication efficiency is more important, Scheme 2 is chosen. This ultimately generates the privacy-efficiency protection optimization strategy for multi-agent pulse communication.

[0155] Furthermore, step S42 includes the following steps:

[0156] Monte Carlo simulation was used in conjunction with privacy-leakage attack scenario models corresponding to pulse sequence reconstruction attack, membership reasoning attack and attribute reasoning attack to simulate and infer pulse attack on randomized sequences of multi-agent pulse perturbation, so as to generate the multi-agent pulse perturbation attack generation process under different attack scenarios.

[0157] In this embodiment of the invention, a privacy-leaking attack scenario model corresponding to Monte Carlo simulation combined with pulse sequence reconstruction attack, membership inference attack, and attribute inference attack is used to simulate pulse attack inference on multi-agent pulse perturbation randomized sequences. Assuming there are 6 agents (A, B, C, D, E, F) in the multi-agent system, and the number of Monte Carlo simulations is set to 10,000, for the pulse sequence reconstruction attack scenario model, in each simulation, the attacker attempts to reconstruct the original pulse sequence based on the received multi-agent pulse perturbation randomized sequence using a specific algorithm (such as a Bayesian inference-based algorithm). Taking agent A as an example, assuming its original pulse sequence is [0, 1, 0, 1, 0], in the simulation, the attacker reconstructs the sequence using the algorithm based on the received perturbation randomized sequence. After multiple simulations, the degree of difference between the reconstruction result and the original sequence is recorded to evaluate the effectiveness of the pulse sequence reconstruction attack. For the membership inference attack scenario model, the attacker attempts to infer whether a certain agent belongs to the multi-agent system by analyzing the pulse sequence. In each simulation, the attacker introduces an unknown agent... The pulse sequence is compared with known multi-agent pulse perturbation randomized sequences (e.g., using a cosine similarity algorithm). Assuming that in a simulation, the pulse sequence of an unknown agent has a similarity of 0.8 with that of agent B, the number of times the unknown agent is identified as belonging to the system is statistically analyzed through multiple simulations to assess the success rate of member inference attacks. For attribute inference attack scenarios, attackers attempt to infer certain attributes of the agent (such as the agent's working state) from the pulse sequence. In each simulation, the attacker uses classification algorithms (such as decision tree classification algorithms) to infer the agent's attributes based on the characteristics of the pulse sequence (such as pulse frequency and pulse amplitude). Assuming that in a simulation, based on the pulse sequence characteristics of agent C, its working state is inferred to be "busy," the number of correct inferences is statistically analyzed through multiple simulations to assess the accuracy of attribute inference attacks. Through 10,000 Monte Carlo simulations, the multi-agent pulse perturbation attack generation process corresponding to different attack scenarios is finally generated. The attack attempts, results, and related parameters in each simulation are recorded in detail, providing a rich data foundation for subsequent analysis.

[0158] Preferably, the pulse decision-making process of the multi-agent pulse perturbation attack generation process under different attack scenarios is subjected to multiple rounds of reverse reasoning to generate a multi-agent pulse perturbation attack decision tree under different attack scenarios.

[0159] In this embodiment of the invention, by performing multi-round reverse reasoning on the pulse decision-making process of the multi-agent pulse perturbation attack generation process under different attack scenarios, taking the pulse sequence reconstruction attack scenario as an example, the reverse reasoning starts from the reconstruction result of each simulation. Assuming that in one simulation, the reconstruction result differs from the original sequence, the reasons for this difference are analyzed, such as the loss or incorrect reconstruction of a pulse. These reasons are used as nodes in the decision tree; for example, node A represents "pulse loss," and node B represents "incorrect pulse reconstruction." Then, the conditions under which these reasons occur are further analyzed. For example, node A can be further divided into sub-nodes A1 representing "noise..." "Audio interference causes pulse loss," and "Algorithm defect causes pulse loss," etc. Through reverse reasoning of multiple simulation results, a complete multi-agent pulse perturbation attack decision tree is gradually constructed. For member reasoning attack scenarios and attribute reasoning attack scenarios, the same method is used for reverse reasoning and decision tree construction. For example, in the member reasoning attack decision tree, node C represents "similarity is higher than the threshold," child node C1 represents "belongs to the system," and C2 represents "does not belong to the system," etc. These decision trees clearly show the logic of attack decision and possible paths under different attack scenarios, providing an intuitive model for subsequent analysis.

[0160] Preferably, the stagnation time points of multi-agent pulse perturbation in different attack scenarios are obtained by using the multi-agent pulse perturbation attack decision tree corresponding to different attack scenarios, and leakage probability prediction calculation is performed on the multi-agent pulse perturbation attack decision tree corresponding to different attack scenarios based on the stagnation time points of multi-agent pulse perturbation in different attack scenarios, so as to generate the multi-agent pulse privacy leakage probability corresponding to different attack scenarios.

[0161] In this embodiment of the invention, by using multi-agent pulse perturbation attack decision trees corresponding to different attack scenarios, the stagnation points of multi-agent pulse perturbation under different attack scenarios are obtained. Taking a pulse sequence reconstruction attack decision tree as an example, the path with the smallest difference between the reconstruction result and the original sequence is found in the tree. The time corresponding to a node on this path is the stagnation point of the pulse perturbation. Assuming that on a certain path, the time corresponding to node D is the 5th simulation, at which point the difference between the reconstruction result and the original sequence is the smallest, then the time point corresponding to the 5th simulation is the stagnation point of the pulse perturbation. Based on the stagnation points of multi-agent pulse perturbation under different attack scenarios, the leakage probability prediction calculation is performed on the multi-agent pulse perturbation attack decision trees corresponding to different attack scenarios. The leakage probability calculation formula is set as P = The formula is n / N, where n is the number of successful privacy leaks after the pulse perturbation stagnation point, and N is the total number of simulations. Taking the pulse sequence reconstruction attack scenario as an example, assuming that the original pulse sequence is successfully reconstructed 100 times after the pulse perturbation stagnation point, and the total number of simulations is 10,000, then the leakage probability P = 100 / 10,000 = 0.01. The same formula is used to calculate the leakage probability for both member inference attack scenarios and attribute inference attack scenarios. For example, in the member inference attack scenario, if the agent member identity is successfully inferred 200 times after the pulse perturbation stagnation point, then the leakage probability P = 200 / 10,000 = 0.02; in the attribute inference attack scenario, if the agent attribute is successfully inferred 150 times, then the leakage probability P = 150 / 10,000 = 0.015. Finally, the multi-agent pulse privacy leakage probabilities corresponding to different attack scenarios are obtained.

[0162] Furthermore, the present invention also provides a multi-agent privacy protection analysis system based on a pulse communication mechanism, for executing the multi-agent privacy protection analysis method based on the pulse communication mechanism described above. The multi-agent privacy protection analysis system based on the pulse communication mechanism includes:

[0163] The spatiotemporal sparse pulse conversion module is used to acquire the continuous input signal sequence corresponding to the multi-agent system, and perform spatiotemporal sparse pulse conversion on the continuous input signal sequence corresponding to the multi-agent system based on the sparse matrix generation method of time coding and rate coding to generate a multi-agent spatiotemporal sparse pulse sequence.

[0164] The pulse communication decision output module is used to construct a hierarchical spatiotemporal spiking neural network and input the multi-agent spatiotemporal sparse pulse sequence into the hierarchical spatiotemporal spiking neural network. Combined with the pulse network training method based on differential privacy, the pulse communication decision output is performed to generate the multi-agent time-series key communication pulse sequence.

[0165] The pulse perturbation randomization module is used to perform pulse perturbation randomization processing on the multi-agent time-critical communication pulse sequence based on a combination of pulse time jitter and pulse deletion, thereby generating a multi-agent pulse perturbation randomized sequence.

[0166] The pulse privacy protection analysis module is used to construct privacy leakage attack scenario models corresponding to pulse sequence reconstruction attacks, membership inference attacks, and attribute inference attacks. Monte Carlo simulation combined with the corresponding privacy leakage attack scenario models is used to simulate privacy leakage attacks on multi-agent pulse perturbation randomized sequences to generate multi-agent pulse privacy leakage probabilities under different attack scenarios. Based on the multi-agent pulse privacy leakage probabilities under different attack scenarios, privacy leakage protection analysis is performed on multi-agent pulse perturbation randomized sequences to generate privacy-efficiency protection optimization strategies for multi-agent pulse communication.

[0167] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0168] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A multi-agent privacy-preserving analysis method of a pulse communication mechanism, characterized in that, The method comprises the following steps: Step S11: acquiring a continuous input signal sequence corresponding to a multi-agent system; Step S12: generating input signal pulses for the continuous input signal sequence corresponding to the multi-agent system to generate a multi-agent input signal pulse sequence; Step S13: converting the multi-agent input signal pulse sequence into a time-delay sparse matrix based on a time coding sparse matrix generation method to generate a multi-agent signal sparse matrix with time-delay characteristics; Step S14: converting the multi-agent input signal pulse sequence into a pulse rate sparse matrix based on a rate coding sparse matrix generation method to generate a multi-agent signal sparse matrix with rate characteristics; Step S15: converting the multi-agent input signal pulse sequence into a multi-agent spatio-temporal sparse pulse sequence based on the multi-agent signal sparse matrix with time-delay characteristics and rate characteristics; Step S2: constructing a hierarchical spatio-temporal pulse neural network, inputting the multi-agent spatio-temporal sparse pulse sequence into the hierarchical spatio-temporal pulse neural network, and outputting a pulse communication decision based on a differential privacy-based pulse network training method to generate a multi-agent time sequence key communication pulse sequence; Step S3: performing pulse perturbation randomization processing on the multi-agent time sequence key communication pulse sequence based on pulse time jitter and pulse deletion to generate a multi-agent pulse perturbation randomization sequence; Step S41: constructing a privacy leakage attack scene model corresponding to a pulse sequence reconstruction attack, a member reasoning attack, and an attribute reasoning attack; Step S42: simulating a privacy leakage attack on the multi-agent pulse perturbation randomization sequence based on the corresponding privacy leakage attack scene model and Monte Carlo simulation to generate a corresponding multi-agent pulse privacy leakage probability under different attack scenes; Step S43: analyzing the privacy leakage diffusion influence of the multi-agent pulse perturbation randomization sequence based on the corresponding multi-agent pulse privacy leakage probability under different attack scenes to obtain a corresponding diffusion path and influence range of a pulse privacy leakage event in the multi-agent system; Step S44: constructing a privacy-efficiency optimization target corresponding to a privacy protection level, a communication bandwidth consumption, and a calculation delay based on the corresponding diffusion path and influence range of the pulse privacy leakage event in the multi-agent system; Step S45: performing privacy leakage protection analysis on the multi-agent pulse perturbation randomization sequence based on the privacy-efficiency optimization target corresponding to the privacy protection level, the communication bandwidth consumption, and the calculation delay, solving a Pareto optimal solution set of the privacy-efficiency optimization target using an NSGA-II algorithm, and obtaining a balance strategy between privacy and efficiency of pulse signal communication based on the Pareto optimal solution set to generate a privacy-efficiency protection optimization strategy corresponding to multi-agent pulse communication.

2. The multi-agent privacy-preserving analysis method of pulse communication mechanism according to claim 1, characterized in that, Step S13 comprises the following steps: Step S131: acquiring a signal pulse time interval and a signal pulse waveform feature corresponding to the multi-agent input signal pulse sequence; Step S132: determining the dilution time window according to the signal pulse time interval and the signal pulse waveform characteristics, to obtain the signal pulse sparse time window size related to the multi-agent original signal time distribution; Step S133: based on the signal pulse sparse time window size related to the multi-agent original signal time distribution, the pulse amplitude and time sequence position of each sparse time window moment in the multi-agent input signal pulse sequence are counted, to obtain the multi-agent signal pulse amplitude and multi-agent signal time sequence position corresponding to each sparse time window moment; Step S134: according to the multi-agent signal pulse amplitude and multi-agent signal time sequence position corresponding to each sparse time window moment, the time delay distribution difference at different moments is calculated, to obtain the time delay distribution difference between multi-agent signal pulses at different sparse time window moments; Step S135: based on the time delay distribution difference between multi-agent signal pulses at different sparse time window moments and combined with the time encoding sparse matrix generation method, the multi-agent input signal pulse sequence is converted into a time delay sparse matrix, to generate a multi-agent signal sparse matrix with time delay characteristics.

3. The multi-agent privacy-preserving analysis method of pulse communication mechanism according to claim 1, wherein, Step S14 includes the following steps: Step S141: the pulse frequency and amplitude in each signal pulse window of the multi-agent input signal pulse sequence are calculated, to obtain the corresponding multi-agent signal pulse frequency and multi-agent signal pulse amplitude in each signal pulse window; Step S142: the corresponding multi-agent pulse signal intensity is obtained through the multi-agent input signal pulse sequence, and based on the multi-agent pulse signal intensity, the pulse rate modulation of the multi-agent input signal pulse sequence is performed to generate the modulation signal pulse rate corresponding to the multi-agent signal pulse; Step S143: based on the modulation signal pulse rate corresponding to the multi-agent signal pulse and combined with the rate encoding sparse matrix generation method, the pulse rate sparse weight reconstruction of the corresponding multi-agent signal pulse frequency and multi-agent signal pulse amplitude in each signal pulse window is performed, to generate a multi-agent signal sparse matrix with rate characteristics.

4. The multi-agent privacy-preserving analysis method of pulse communication mechanism according to claim 1, wherein, Step S15 includes the following steps: Step S151: based on the multi-agent signal sparse matrix with time delay characteristics, the multi-agent input signal pulse sequence is filtered in time to obtain a multi-agent time delay sparse filtered pulse sequence; Step S152: based on the multi-agent signal sparse matrix with rate characteristics, the multi-agent time delay sparse filtered pulse sequence is modulated in spatial position distribution to generate a multi-agent rate sparse modulated pulse sequence; Step S153: the pulse energy size corresponding to each pulse signal is obtained through the multi-agent rate sparse modulated pulse sequence, and the pulse attenuation evaluation is performed according to the pulse energy size corresponding to each pulse signal, to obtain the pulse attenuation factor corresponding to the multi-agent pulse signal; Step S154: based on the pulse attenuation factor corresponding to the multi-agent pulse signal, the corresponding each pulse signal in the multi-agent rate sparse modulated pulse sequence is optimized in time and space pulse distribution, to generate a multi-agent time and space sparse pulse sequence.

5. The multi-agent privacy-preserving analysis method of pulse communication mechanism according to claim 1, wherein, Step S2 includes the following steps: Step S21: Constructing a corresponding hierarchical spatiotemporal pulse neural network by adopting a structure combining a recurrent neural network and a long short-term memory network, specifically including a bottom layer, a middle layer and a top layer, wherein the bottom layer is used to process the original pulse sequence, the middle layer is used to extract the corresponding pulse timing feature, and the top layer is used to output the corresponding communication pulse sequence; Step S22: Inputting the multi-agent spatiotemporal sparse pulse sequence into the bottom layer of the hierarchical spatiotemporal pulse neural network for pulse preprocessing, and inputting the preprocessed multi-agent spatiotemporal sparse pulse sequence into the middle layer of the hierarchical spatiotemporal pulse neural network for pulse extraction network training by introducing a pulse timing attention mechanism, while designing and constructing a pulse network training method based on differential privacy in the training process to introduce Gaussian noise into the pulse neural network, and combining a gradient clipping strategy in the training process to prevent gradient reversal attacks, to train the generation of pulse timing feature outputs of each agent; Step S23: The pulse interaction frequency and the pulse sensitivity between the agents are calculated by the pulse time sequence feature output corresponding to each agent at the top layer in the layered spatiotemporal pulse neural network, and the corresponding privacy budget is dynamically configured according to the pulse interaction frequency and the pulse sensitivity between the agents parameters, and the privacy budget parameters are used to decide the output of the corresponding communication pulse sequence in the top layer, and the multi-agent time sequence key communication pulse sequence is generated.

6. The multi-agent privacy-preserving analysis method of pulse communication mechanism according to claim 1, wherein, Step S3 includes the following steps: Step S31: Pulse timing disturbance analysis is performed on the multi-agent timing key communication pulse sequence to obtain multi-agent communication pulse timing disturbance distribution characteristics; Step S32: The pulse communication disturbance delay and the pulse communication disturbance frequency difference between each agent are obtained through the multi-agent communication pulse timing disturbance distribution characteristics, and the pulse time jitter amplitude of each pulse signal in the multi-agent timing key communication pulse sequence is statistically obtained based on the pulse communication disturbance delay and the pulse communication disturbance frequency difference between each agent, to obtain the communication pulse signal time jitter amplitude between each agent; Step S33: A corresponding multi-agent pulse time jitter amplitude atlas is generated according to the communication pulse signal time jitter amplitude between each agent, and a pulse selective deletion optimization is performed on each pulse signal in the multi-agent timing key communication pulse sequence based on the multi-agent pulse time jitter amplitude atlas, to screen and delete the redundant pulse signals corresponding to the consistent communication pulse signal time jitter amplitude, to generate a multi-agent pulse optimization deletion sequence; Step S34: Pulse disturbance distribution randomization processing is performed on the timing of the corresponding remaining pulse signals in the multi-agent pulse optimization deletion sequence to generate a multi-agent pulse disturbance randomization sequence.

7. The multi-agent privacy-preserving analysis method of pulse communication mechanism according to claim 1, wherein, Step S42 includes the following steps: Monte Carlo simulation is combined with the corresponding privacy leakage attack scene model of pulse sequence reconstruction attack, member inference attack and attribute inference attack to perform pulse attack simulation reasoning on the multi-agent pulse disturbance randomization sequence to generate the corresponding multi-agent pulse disturbance attack generation process under different attack scenes; Pulse decision-making multi-round reverse reasoning is performed on the corresponding multi-agent pulse disturbance attack generation process under different attack scenes to generate the corresponding multi-agent pulse disturbance attack decision tree under different attack scenes; The corresponding multi-agent pulse disturbance attack decision tree under different attack scenarios is obtained by corresponding multi-agent pulse disturbance attack decision tree under different attack scenarios. For example, in the tree, find the path with the smallest difference between the reconstruction result and the original sequence. The time corresponding to a node on this path is the multi-agent pulse disturbance stall time point. Based on the corresponding multi-agent pulse disturbance stall time point under different attack scenarios, the corresponding multi-agent pulse disturbance attack decision tree under different attack scenarios is calculated to generate the corresponding multi-agent pulse privacy leakage probability under different attack scenarios.

8. A multi-agent privacy-preserving analysis system of a pulse communication mechanism, characterized in that, A multi-agent privacy protection analysis method for implementing the pulse communication mechanism of claim 1, the pulse communication mechanism of multi-agent privacy protection analysis system comprises: A space-time sparse pulse conversion module is used to obtain a continuous input signal sequence corresponding to a multi-agent system, and based on a time encoding and rate encoding sparse matrix generation method, the continuous input signal sequence corresponding to the multi-agent system is converted into a space-time sparse pulse sequence by space-time sparse pulse conversion. A pulse communication decision output module is used to construct a hierarchical space-time pulse neural network, and input the multi-agent space-time sparse pulse sequence into the hierarchical space-time pulse neural network to output pulse communication decision based on differential privacy-based pulse network training method, thereby generating a multi-agent time sequence key communication pulse sequence. A pulse disturbance randomization module is used to perform pulse disturbance randomization processing on the multi-agent time sequence key communication pulse sequence based on pulse time jitter and pulse deletion combination, thereby generating a multi-agent pulse disturbance randomization sequence. A pulse privacy protection analysis module is used to construct a privacy leakage attack scenario model corresponding to pulse sequence reconstruction attack, member reasoning attack and attribute reasoning attack, and to simulate the privacy leakage attack on the multi-agent pulse disturbance randomization sequence based on the corresponding privacy leakage attack scenario model by Monte Carlo simulation, thereby generating the corresponding multi-agent pulse privacy leakage probability under different attack scenarios. Based on the corresponding multi-agent pulse privacy leakage probability under different attack scenarios, the privacy leakage protection analysis of the multi-agent pulse disturbance randomization sequence is performed to generate the privacy-efficiency protection optimization strategy corresponding to the multi-agent pulse communication.

Images