Concentrator network switching decision-making system and method based on communication quality evaluation

Abstract

The invention discloses a concentrator network switching decision-making system and method based on communication quality evaluation. Safe and intelligent switching is achieved through cooperation of multiple modules. Constructing a physical, network and application layer multi-dimensional index system, and generating a communication quality score by adopting a dynamic weight and a random forest; the safety protection module ensures signal legality through radio frequency fingerprint and digital signature two-factor authentication; the switching decision is based on a double-threshold mechanism, trend prediction and a pre-switching mechanism, so that the interruption time is shortened; and the privacy protection adopts dynamic differential privacy and federated learning to realize data desensitization and model security aggregation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of wisdom communication, and particularly relates to a concentrator network switching decision system and method based on communication quality evaluation. BACKGROUND

[0002] In the Internet of Things and wireless sensor networks, the concentrator as a data aggregation node needs to perform network switching in a multi-access point environment to ensure communication quality. In the prior art, the security and privacy protection of the switching decision mainly adopts the following schemes: 1. Data encryption technology: the communication quality indicators (such as RSSI, SNR) are transmitted after traditional encryption (such as AES), to prevent data from being stolen during transmission. 2. Signal strength authentication: the legality of the candidate network is judged by verifying the signal strength threshold and simple beacon frame check. In the above schemes, the most similar prior art is the privacy protection based on a single encryption means and the node authentication based on signal strength. It protects privacy by encrypting transmitted data, and authenticates the legality of nodes by presetting signal strength threshold and simple check code, to realize basic security protection of the switching decision process.

[0003] The prior art still has the following deficiencies: 1. Incomplete privacy protection: only the transmitted data is encrypted, and the features of the data itself (such as space-time trajectory, device identity) are not desensitized, so attackers can infer sensitive information by analyzing the statistical characteristics or trajectory patterns of encrypted data, which poses a risk of privacy leakage. 2. Weak anti-forgery attack capability: relying on a single physical layer indicator such as signal strength for node authentication, without verifying the hardware inherent characteristics (such as radio frequency fingerprint) of the signal source, malicious nodes can easily induce the concentrator to switch to an illegal network by forging high signal strength, resulting in a high attack success rate. 3. Insufficient cross-layer cooperation: the security protection and decision mechanism are independent of each other, without forming a linkage between physical layer feature authentication and decision layer trust evaluation, which cannot dynamically adapt to complex attack scenarios, resulting in high false negative rate and false positive rate. 4. Poor resource adaptability: the traditional encryption algorithm and authentication mechanism do not consider the power and energy consumption limitations of low-power concentrators, which may cause the battery life to be shortened or the processing delay to be increased when running on embedded devices. 5. Single communication quality evaluation: the existing scheme lacks a multi-dimensional communication quality evaluation system, relying only on a single indicator such as signal strength, which cannot comprehensively evaluate complex indicators at the physical layer, network layer and application layer, resulting in low switching accuracy. SUMMARY

[0004] Therefore, it is necessary to provide a concentrator network switching decision system and method based on communication quality evaluation to realize safe and efficient network switching through multi-module cooperation.

[0005] In a first aspect, the application provides a concentrator network switching decision system based on communication quality evaluation, as shown in Figure 1 The system comprises: A communication quality evaluation module for collecting multi-dimensional indicators of physical layer, network layer and application layer, and generating a communication quality score Q through preprocessing, dynamic weight allocation and comprehensive scoring model. A security protection module including a radio frequency fingerprint recognition unit and a digital signature verification unit for verifying the legitimacy of candidate network signal sources. A switching decision module for generating a switching decision based on the communication quality score Q and the signal source legitimacy verification result, combined with a double threshold mechanism. A privacy protection module for implementing privacy protection of switching decision data through dynamic differential privacy processing and federated learning model. The communication quality evaluation module interacts with the security protection module, switching decision module and privacy protection module in both directions, forming a security protection and decision closed loop with cross-layer cooperation.

[0006] In one embodiment, the multi-dimensional indicators include: Physical layer indicators: received signal strength (RSSI), signal-to-noise ratio (SNR), bit error rate (BER), and multipath fading coefficient. Network layer indicators: transmission delay, throughput, packet loss rate, and link stability. Application layer indicators: service type priority, data integrity requirement, and battery remaining capacity.

[0007] In one embodiment, the communication quality evaluation module includes: A preprocessing unit for denoising time-varying indicators using Kalman filtering and extracting a 50-dimensional feature vector containing time series features. A dynamic weight allocation unit for constructing an indicator weight matrix based on improved analytic hierarchy process (AHP) and entropy weight method, and triggering weight recalculation when the indicator information entropy changes by more than 20%. A comprehensive scoring unit for outputting the communication quality score Q using a lightweight random forest model, and setting a score correction factor (real-time service x1.2, non-real-time service x0.8) according to service type.

[0008] In one embodiment, the security protection module includes: The radio frequency fingerprint recognition unit extracts the carrier frequency offset, phase noise and other hardware inherent characteristics of the signal through the CNN-LSTM network, and performs cosine similarity matching (threshold >= 0.85) with the legal AP feature library. The digital signature verification unit verifies the ECC signature in the beacon frame, and the signature content includes a timestamp (validity period <= 50ms) and a random number (nonce) to prevent replay attacks.

[0009] In one embodiment, the switching decision module includes: The dynamic trust degree calculation unit is based on the formula Trust degree = 0.35 x radio frequency fingerprint matching degree + 0.3 x signature verification result + 0.25 x Q + 0.1 x historical switching success rate Generate the trust degree of the signal source; The dual-threshold triggering unit sets a security threshold (radio frequency fingerprint matching degree > 0.85 and signature verification passes) and a quality threshold (candidate network Q > current network Q + 15 points and Q > 70 points) to jointly trigger switching; The pre-switching unit predicts the quality trend in the future 200ms based on the ARIMA model, and starts pre-switching 50ms in advance when the predicted current network Q < 50 points and the candidate network Q > 80 points.

[0010] In one embodiment, the privacy protection module includes: The dynamic differential privacy unit injects Laplace noise into the spatiotemporal trajectory data, and the noise standard deviation is adaptively adjusted according to the mobile speed (high-speed scene >= low-speed scene 2 times); The federated learning unit uses a hierarchical architecture to aggregate local model gradient parameters at the edge node, and the cloud aggregates globally through secure multi-party computation (MPC), and adds Gaussian noise (sigma = 0.1) when updating the gradient.

[0011] In a second aspect, the application provides a concentrator network switching decision method based on communication quality evaluation, including the steps of: S1. Multi-dimensional index collection and preprocessing: real-time collection of physical layer, network layer and application layer indicators, extraction of time series feature vectors containing the first 4 cycles of data after Kalman filter denoising; S2. Communication quality comprehensive evaluation: dynamically allocate index weights by improving the AHP-entropy weight method, generate communication quality score Q using the random forest model, and correct the score according to the business type; S3. Signal source legality verification: simultaneously perform radio frequency fingerprint matching (cosine similarity >= 0.85) and digital signature verification (ECC signature containing timestamp and nonce); S4. Cross-layer cooperative handover decision: based on the communication quality score Q and the signal source verification result, a handover decision is generated through a double threshold mechanism (safety threshold + quality threshold), triggering pre-handover resource reservation or maintaining the current connection; S5. Privacy protection processing: dynamic differential privacy processing is performed on the collected data, and model parameter safe aggregation is realized through federated learning.

[0012] In one embodiment, the dynamic allocation of index weights includes: A three-layer index system (target layer - rule layer - index layer) is constructed, and the weights of the physical layer, network layer and application layer are 0.4, 0.35 and 0.25 respectively. When the information entropy of any index layer changes by more than 20%, the subjective weight of AHP is corrected based on the entropy weight method.

[0013] In one embodiment, the double threshold mechanism specifically includes: Only when the candidate network meets (1) radio fingerprint matching degree > 0.85 and digital signature verification passes; (2) candidate network Q > current network Q + 15 points and Q > 70 points at the same time, the handover decision is triggered.

[0014] In one embodiment, the dynamic differential privacy processing includes: For high-speed scenarios with a mobile speed > 30km / h, the standard deviation of Laplace noise is increased to 2 times that of low-speed scenarios; the privacy protection parameter ε is dynamically adjusted through an exponential mechanism.

[0015] This application integrates dynamic differential privacy and federated learning through a two-layer privacy protection mechanism, achieving data feature desensitization and secure aggregation of model parameters, protecting the spatiotemporal trajectory privacy of switching decision data. A cross-layer signal source legitimacy authentication system is constructed by combining two-factor signal source authentication with physical-layer feature recognition of RF fingerprints and cryptographic verification of digital signatures. A dynamic adaptive strategy dynamically adjusts privacy protection parameters and security detection thresholds based on scenario requirements, balancing the privacy and security requirements of different services. Lightweight hardware acceleration uses FPGAs for real-time RF fingerprint extraction, and model quantization technology is combined with low-power concentrators to reduce computational overhead. Based on a three-level communication quality assessment system, a 20+-dimensional indicator system covering the physical, network, and application layers is constructed to address the single-metric issue of traditional solutions. The innovative integration of AHP-entropy weight method dynamic weight allocation and a lightweight random forest scoring model achieves real-time assessment within 0.5ms. Through cross-module linkage of quality scoring and trust calculation, a closed-loop control system of "security authentication-quality assessment-intelligent decision-making" is formed. Compared with existing technologies, this system offers more comprehensive privacy protection. It not only encrypts transmitted data but also desensitizes raw features through differential privacy and protects model parameters through federated learning, preventing statistical attacks and trajectory reconstruction, thereby reducing the risk of privacy leakage by 85%. It also offers enhanced attack resistance: its two-factor authentication mechanism achieves a 99.2% malicious signal recognition rate, minimizing false alarm rates, effectively defending against forged signal attacks, and improving handover success rates. Cross-layer collaboration is more intelligent, with physical layer authentication and quality assessment linked to decision-making layer assessments. Signal source trust is dynamically updated, enabling real-time adaptation to complex attack scenarios and reducing misjudgment rates. Decision accuracy is significantly improved, with a newly added communication quality assessment module that comprehensively considers both signal legitimacy and transmission quality in handover decisions. This significantly increases the correct handover rate compared to existing solutions (which rely solely on signal strength or single security authentication). Enhanced adaptability to complex scenarios, combined with dynamic weight allocation and service type awareness, enables the system to achieve decision adaptation times of less than 10ms in both real-time industrial control (requiring low latency) and massive data acquisition (requiring high throughput), meeting the differentiated needs of different industries. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 A schematic diagram of the system architecture of an embodiment of the present application; Figure 2 Schematic diagram of the internal structure of a computer device in one embodiment of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0019] The present application provides a concentrator network switching decision system based on communication quality evaluation, comprising: A communication quality evaluation module is configured to collect multi-dimensional indicators of physical layer, network layer and application layer, and generate a communication quality score Q through pre-processing, dynamic weight distribution and a comprehensive scoring model. A security protection module includes a radio frequency fingerprint recognition unit and a digital signature verification unit, configured to verify the legitimacy of the candidate network signal source. A switching decision module is configured to generate a switching decision based on the communication quality score Q and the signal source legitimacy verification result, combined with a double threshold mechanism. A privacy protection module is configured to realize privacy protection of switching decision data through dynamic differential privacy processing and a federated learning model. The communication quality evaluation module, the security protection module, the switching decision module and the privacy protection module interact with each other to form a security protection and decision closed loop with cross-layer cooperation.

[0020] In one embodiment, the multi-dimensional indicators include: Physical layer indicators: received signal strength (RSSI), signal-to-noise ratio (SNR), bit error rate (BER), multipath fading coefficient; Network layer indicators: transmission delay, throughput, packet loss rate, link stability; Application layer indicators: service type priority, data integrity requirement, battery remaining capacity.

[0021] The received signal strength is measured by the radio frequency front end of the wireless chip to measure the power strength of the received signal, with the unit of dBm, reflecting the strength of the signal. For example, the RSSI sampling accuracy of TI CC2640R2F chip is ±1dBm, which can be used to judge the signal coverage quality of the current link.

[0022] The signal-to-noise ratio is the ratio of signal power to noise power (unit: dB), which is calculated by FFT (Fast Fourier Transform) for frequency spectrum analysis of the received signal. The higher the SNR, the better the signal quality.

[0023] The bit error rate (BER) is the ratio of the number of error bits to the total number of bits in a transmission, obtained through MAC layer CRC (Cyclic Redundancy Check) error statistics. For example, if one error occurs for every 1000 bits transmitted, the BER is 10⁻³.

[0024] The multipath fading coefficient is the amplitude and phase changes caused by wireless signals reaching the receiving end through multiple paths. It uses pilot symbols in the OFDM (Orthogonal Frequency Division Multiplexing) system to perform channel estimation and reflects the time-varying characteristics of the channel.

[0025] Transmission delay: The round-trip time (RTT) is measured by sending a probe packet via ICMP Echo Request with an accuracy of ±5ms, reflecting the real-time nature of network transmission.

[0026] The throughput is the amount of data successfully transmitted per unit time, calculated by counting UDP packets (e.g., if 10 1024-byte packets are transmitted per second, the throughput is 81.92 kbps).

[0027] Packet loss rate: The ratio of IP layer packet loss to the total number of sent packets, used to measure network stability.

[0028] The link stability is determined by the continuous connection time (if there is no interruption within 300ms, it is considered stable), which reflects the continuity of the network connection.

[0029] The service type priority is to identify the service type through a Diffserv (Differentiated Services) tag (0-63), such as a real-time control service (tag value ≥ 48) taking precedence over a non-real-time collection service.

[0030] The data integrity requirement is as follows: the accuracy of data transmission is measured by a CRC check pass rate (e.g., ≥99% is a high requirement).

[0031] The remaining battery capacity is obtained by sampling the battery voltage through an ADC and mapping it to a capacity percentage (eg, 3.7V corresponds to 70% capacity) for energy consumption optimization decisions.

[0032] In one embodiment, the communication quality assessment module further includes: The pre-processing unit uses Kalman filtering to denoise the time-varying indicators and extract a 50-dimensional feature vector containing time series characteristics; The 50-dimensional time series feature extraction extracts 5 types of features (mean, variance, maximum, minimum, trend item) of the current time and the data of the previous 4 periods (200ms) by using a 50ms sliding window, 10-dimensional features per period, a total of 50 dimensions. For example, for the SNR sequence, the mean value of the SNR in the window is extracted: the average value of the signal strength reflects the overall level; The variance measures the fluctuation of the SNR, and the larger the variance, the more unstable the channel; The trend item is calculated by linear regression to calculate the slope (such as SNR decreasing by 2dB per second, indicating signal deterioration).

[0033] In one embodiment, the communication quality evaluation module further comprises: A dynamic weight allocation unit based on improved analytic hierarchy process (AHP) and entropy weight method constructs an index weight matrix, and triggers the recalculation of the weight when the index information entropy changes by more than 20%; The analytic hierarchy process first constructs a three-level index system, the target layer: communication quality score Q; The criterion layer: physical layer (0.4), network layer (0.35), application layer (0.25); The index layer: the physical layer includes RSSI (0.3), SNR (0.3), BER (0.4), etc.

[0034] Then construct the judgment matrix A: take the physical layer index (RSSI, SNR, BER) as an example, A =

[0035] Among them, A 13 =1 / 2: indicates that RSSI is slightly less important than BER (because BER directly reflects transmission errors, and RSSI only reflects signal strength); A 31 =2: BER is slightly more important than RSSI, and A 13 The reciprocal of each other; A 12 =1: RSSI and SNR are equally important (both are basic indicators of signal quality).

[0036] Then calculate the maximum eigenvalue and eigenvector of the judgment matrix A: after matrix column normalization and row sum normalization, the maximum eigenvalue and eigenvector are calculated by power method or eigenvalue decomposition, and the weight vector w AHP =[0.2, 0.2, 0.6] after normalization, that is, the weight of BER is 0.6 (higher than RSSI and SNR).

[0037] The entropy weight method dynamically corrects the weight matrix, and objectively measures the importance of indicators through the information entropy of indicators. The steps are as follows: Calculate the information entropy: , where is the ith standardized value of indicator j , and n is the number of samples.

[0038] Calculate the information utility value: The larger the utility value, the more important the indicator.

[0039] Correct the weight: For example, when the information entropy of the packet loss rate changes by more than 20% (such as network congestion), its weight will be increased from 0.3 to 0.4, strengthening the influence on the switching decision.

[0040] The entropy weight method measures the "discrimination" of indicators through the information entropy of indicators: If the difference between indicator values is large (such as large fluctuations in packet loss rate during sudden congestion), the information entropy is low, the discrimination is high, and a higher weight should be given; if the indicator values tend to be consistent (such as small changes in RSSI in a stable channel), the information entropy is high, the discrimination is low, and the weight should be reduced.

[0041] Taking network layer indicators as an example, assume that the network layer indicators are delay, throughput, and packet loss rate, and the standardized data collected at a certain time is shown in Table 1: Table 1 Standardized network layer indicator collection table

[0042] Calculate the probability matrix:

[0043] Taking the packet loss rate as an example,

[0044] p 13 =1.5 / 0.1≈0.067, p 23 =1.5 / 0.9=0.6, p 33 =1.5 / 0.5≈0.333 Information entropy of packet loss rate:

[0045] Information utility value of packet loss rate:

[0046] Assume that the initial weight of AHP is w AHP =[0.4,0.3,0.3], then the corrected weight of the entropy weight method is: , calculate the denominator: ; Assume , then ; The packet loss rate has the most weight: ; Compared with the initial weight of 0.3, the revised weight is increased by 57%, reflecting the increase in the importance of the packet loss rate in the current network congestion scenario.

[0047] When the information entropy of any indicator changes by more than 20% (such as the information entropy of the packet loss rate decreasing from 0.8 to 0.64), the weight is recalculated.

[0048] When the service type is switched from non-real-time (such as data collection) to real-time (such as industrial control), the preset judgment matrix is automatically loaded.

[0049] In one embodiment, the communication quality assessment module further includes: A comprehensive scoring unit outputs a communication quality score Q using a lightweight random forest model, and sets a score correction factor (real-time service × 1.2, non-real-time service × 0.8) in combination with the service type.

[0050] The random forest algorithm constructs multiple decision trees and outputs the result by voting, and has anti-noise and overfitting capabilities. In this module, the parameters of the random forest are set as follows: tree depth ≤ 8, node number ≤ 32, and the optimal splitting feature is selected by the Gini index (measuring the purity improvement after feature splitting), with the formula being: , where p k is the proportion of the kth class of samples in the node.

[0051] The service type correction factor setting includes: Real-time service (such as industrial control): score correction factor × 1.2, to strengthen the sensitivity of delay / packet loss rate. For example, a candidate network Q=70 points, corrected to 84 points under real-time service, the probability of triggering switching is higher.

[0052] Non-real-time service (such as data collection): score correction factor × 0.8, to reduce the impact of transient fluctuations. For example, Q=65 points corrected to 52 points, to avoid false switching due to short-term noise.

[0053] In one embodiment, the security protection module includes: A radio frequency fingerprint recognition unit extracts hardware inherent features such as carrier frequency offset and phase noise of the signal through a CNN-LSTM network, and performs cosine similarity matching (threshold ≥ 0.85) with the legal AP feature library; The digital signature verification unit verifies the ECC signature in the beacon frame, and the signature content includes a time stamp (validity period ≤ 50 ms) and a random number (nonce) to prevent replay attacks.

[0054] The radio frequency fingerprinting unit: based on the physical layer authentication of hardware features, the identity of the signal source is realized by extracting the hardware inherent characteristics of the wireless signal, and the real-time feature analysis and matching are completed by using FPGA hardware acceleration. A real-time processing platform is constructed by using Xilinx Zynq Ultrascale+ FPGA, and a 2.4GHz frequency band signal is collected by integrating an analog front end (AD9361 radio frequency chip), and converted into baseband IQ data through digital down conversion (DDC). The CNN-LSTM neural network architecture is used to extract signal features, including carrier frequency offset, phase noise, amplifier non-linear distortion and other hardware inherent characteristics, to form a 100-dimensional feature vector; the feature template library (100-dimensional features per template) of the legal AP is pre-stored, and the matching degree of the real-time signal features and the template is calculated by using the cosine similarity algorithm, and the matching threshold is set to 0.85, and only when the similarity exceeds the threshold, it is determined as a legal signal source; the time window mechanism is used to update the feature library periodically, and the feature parameter drift tolerance of ±5% is allowed for the AP used for a long time, which adapts to the feature change caused by hardware aging and reduces the false recognition rate.

[0055] The digital signature verification unit: based on the message integrity authentication of cryptography, the identity of the signal source and the message integrity are verified by using the ECC digital signature technology to prevent fake signals and replay attacks. The P-256 elliptic curve cryptography algorithm is used, the AP generates the ECC signature of the beacon frame content (SNR, channel number, time stamp, random number nonce), and the signature validity period is limited to 50ms (the time stamp and the clock difference of the receiving end ≤ 50ms). After the concentrator receives the beacon, the time stamp validity is verified first, then the message hash value is calculated by using the SHA-256 algorithm, and the pre-stored AP public key is used to verify the signature parameters (r, s), to ensure that the message is not tampered and the source is legal. Each beacon frame contains a 64-bit random number nonce, combined with the time stamp mechanism, which effectively resists replay attacks (repeatedly sent old beacons are rejected due to expired time stamp or repeated nonce).

[0056] In one embodiment, the security protection module further comprises: The dual-factor authentication cooperation unit, based on a cross-layer cooperation architecture, first performs a parallel verification process, simultaneously starting radio frequency fingerprint matching and digital signature verification on the candidate network signal, and only when both pass (matching degree > 0.85 and signature valid) triggers the subsequent switching process, forming a double protection of physical layer features and password layer verification; then, dynamic trust modeling is performed: the verification result is integrated into the trust calculation model, the radio frequency matching degree (0-1) and the signature result (0 / 1) respectively affect the trust degree with a weight of 40% and 30%, combined with the communication quality score and the historical switching record, to realize dynamic trust evaluation of the signal source; finally, an abnormal response strategy is generated: for the signal source that fails 5 times in a row, a temporary blacklist mechanism is started (banned for 10 minutes), and for the abnormal situation that the matching degree meets the standard but the signature fails, the sampling rate is automatically increased to 20 MSPS for deep feature analysis to identify hardware feature forgery attacks.

[0057] In one embodiment, the switching decision module includes: The dynamic trust calculation unit is used for multi-dimensional information fusion evaluation, which quantifies the multi-dimensional information such as communication quality score, signal source legitimacy, and historical switching record into trust degree index as the core basis for switching decision.

[0058] Evaluation dimensions and weight distribution: Radio frequency fingerprint matching degree (0-1, weight 35%): the similarity quantization result based on physical layer features; Digital signature verification result (0 / 1, weight 30%): password layer authentication pass; Communication quality score Q (0-100, normalized weight 25%): comprehensive quality score from the evaluation module; Historical switching success rate (0-1, weight 10%): the success rate of the last 100 switching.

[0059] The trust degree quantization model uses a linear weighted model to calculate the trust degree based on the formula TrustScore = 0.35 x Radio frequency fingerprint matching degree +0.3× Signature verification result +0.25×(Q / 100) +0.1 × Historical switching success rate Generate the trust degree of the signal source. When the trust degree is greater than or equal to 0.7, the candidate network is determined to be trusted.

[0060] In one embodiment, the switching decision module includes: The double threshold triggering unit sets a security threshold (radio fingerprint matching degree > 0.85 and signature verification passes) and a quality threshold (candidate network Q > current network Q + 15 points and Q > 70 points) to jointly trigger the switching; the security threshold ensures the physical layer legality and message integrity of the candidate network, preventing malicious nodes from accessing; the quality threshold avoids switching to a legitimate network with insignificant quality improvement or low quality itself, reducing invalid switching, wherein the candidate network Q value > current network Q value + 15 points is a quality gain requirement, and the candidate network Q value > 70 points is an absolute quality threshold. Only when the security threshold and the quality threshold are met at the same time, the switching process is started, and compared with a single threshold, the false switching rate is reduced by 45%.

[0061] In one embodiment, the switching decision module includes: The pre-switching unit predicts the future 200ms quality trend based on the ARIMA model, and starts the pre-switching 50ms in advance when the predicted current network Q < 50 points and the candidate network Q > 80 points.

[0062] The pre-switching unit makes an early decision based on trend prediction, uses an ARIMA (p, d, q) time series model to predict the Q value change in the future 200ms, and the typical parameter configuration is: autoregressive order p = 2, difference order d = 1, and moving average order q = 1; the historical data window: store the Q value sequence of the last 100 periods (5 seconds).

[0063] The pre-switching triggering condition is to start the pre-switching 50ms in advance when the prediction meets the following conditions: The predicted current network Q value < 50 points (quality deterioration warning); the predicted candidate network Q value > 80 points (high quality guarantee).

[0064] The pre-switching resource reservation process is: first, send a pre-authentication request (carrying the current connection context) to the candidate network; Then the candidate network reserves bandwidth resources according to the QoS requirement (such as allocating a fixed priority queue for real-time services); Finally, the concentrator starts the data buffering mechanism, and guarantees data continuity through the retransmission mechanism when switching, so that the switching interruption time is < 8ms.

[0065] In one embodiment, the switching decision module includes: The decision optimization unit makes decision optimization by designing a hysteresis threshold and perceiving a global load. The hysteresis threshold design is to set a switching hysteresis interval [Q_low, Q_high], for example, after switching from network A to network B, if the Q value of A rises to Q_current+10 within a certain range, it will not switch back immediately, avoiding the ping-pong effect caused by signal fluctuation. Global load perception is when multiple concentrators switch to the same candidate network at the same time, the decision module obtains the network load state (such as the number of AP access users) through the edge node, if the load rate > 70%, the quality score of the network is reduced (multiplied by a decay coefficient of 0.8), guiding load balancing.

[0066] The module realizes the whole-process intelligent decision of "legal network screening - high-quality network optimization - future quality prediction" through a three-layer decision mechanism of trust quantification, double-threshold cooperation and trend prediction, and is especially suitable for industrial Internet of Things, Internet of Vehicles and other scenes with strict requirements on reliability and real-time performance.

[0067] In one embodiment, the privacy protection module includes: The dynamic differential privacy unit injects Laplace noise into the spatiotemporal trajectory data, and the noise standard deviation is adaptively adjusted according to the moving speed (high-speed scene ≥ low-speed scene 2 times); The dynamic differential privacy unit realizes privacy protection of switching decision data through adaptive noise injection based on the noise desensitization mechanism of spatiotemporal data, preventing sensitive information such as spatiotemporal trajectory and device state from being reconstructed.

[0068] Laplace mechanism is used to add noise to sensitive data (such as position coordinates and switching timestamps), and adaptive moving speed is considered, with the noise standard deviation in high-speed scenes being increased to 2 times that in low-speed scenes, enhancing the trajectory confusion effect; the privacy parameter is dynamically adjusted by an exponential function, and the spatiotemporal trajectory is blurred, with k-anonymity (k=10) being used in the spatial dimension to map the real coordinates to a virtual grid, ensuring at least 10 trajectory points in each grid; the switching timestamp is discretized to a granularity of 50ms in the time dimension, destroying the spatiotemporal correlation characteristics.

[0069] The federated learning unit adopts a hierarchical architecture to aggregate local model gradient parameters at the edge node, and the cloud aggregates globally through secure multi-party computation (MPC), adding Gaussian noise (σ=0.1) when updating the gradient.

[0070] The federated learning unit, based on the secure aggregation architecture of model parameters, realizes cross-node model collaborative update through hierarchical federated learning, ensuring that the original data does not leave the local and only encrypted gradient parameters are uploaded.

[0071] The hierarchical learning architecture design is divided into an edge node layer and a cloud layer. The local model gradient of 100 concentrators in the region is aggregated on the edge node layer, and distributed training is realized using the ClientSet API of TensorFlow Federated. On the cloud layer, global model aggregation is performed through secure multi-party computation (MPC, such as the ABY3 protocol) to avoid the risk of data leakage of the central node. Gaussian noise is added to the gradient during local training, and gradient clipping (ClipGradients) is used to limit the influence of abnormal gradients and prevent the original data features from being obtained through gradient backpropagation. The local model uses a 2-layer fully connected neural network (64 units per layer), and the model size is compressed to less than 1MB through 8-bit quantization (INT8). The training period is set to 10 minutes per round, and the energy consumption of single aggregation is controlled to be less than 5mW.

[0072] Through the cooperation of dynamic noise injection and distributed model training, the block realizes the full-life-cycle privacy protection of switching decision data without sacrificing model accuracy, and is especially suitable for critical infrastructure fields with strict requirements for data security and privacy.

[0073] In a second aspect, the application provides a concentrator network switching decision method based on communication quality evaluation, comprising the steps of: S1. Multi-dimensional index collection and preprocessing: real-time collection of physical layer, network layer and application layer indexes, extraction of time series feature vectors containing the first 4 cycle data after Kalman filter denoising; S2. Communication quality comprehensive evaluation: dynamically allocate index weights by improving the AHP-entropy weight method, generate communication quality score Q using the random forest model, and correct the score according to the business type; S3. Signal source legality verification: synchronous radio frequency fingerprint matching (cosine similarity ≥ 0.85) and digital signature verification (ECC signature containing timestamp and nonce); S4. Cross-layer collaborative switching decision: based on the communication quality score Q and the signal source verification result, generate a switching decision through a double-threshold mechanism (security threshold + quality threshold), trigger pre-switching resource reservation or maintain the current connection; S5. Privacy protection processing: dynamic differential privacy processing is performed on the collected data, and model parameter safe aggregation is realized through federated learning.

[0074] In one embodiment, the dynamic allocation of index weights includes: A three-layer index system (target layer - criterion layer - index layer) is constructed, and the weights of the physical layer, network layer and application layer are 0.4, 0.35 and 0.25 respectively. When the information entropy of any index layer changes by more than 20%, the subjective weight of AHP is corrected based on the entropy weight method.

[0075] In one embodiment, the dual threshold mechanism specifically includes: Only when the candidate network meets (1) the radio fingerprint matching degree > 0.85 and the digital signature verification passes; (2) the candidate network Q > the current network Q + 15 points and Q > 70 points at the same time, the switching decision is triggered.

[0076] In one embodiment, the dynamic differential privacy processing includes: For high-speed scenarios with a moving speed > 30km / h, the standard deviation of Laplace noise is increased to 2 times of that in low-speed scenarios; the privacy protection parameter is dynamically adjusted through an exponential mechanism.

[0077] In one example embodiment, a computer device, which can be a terminal, is provided, and an internal structure diagram of the computer device can be as shown in Figure 2 The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by the processor to implement a surgical instrument life calculation method and a surgical instrument control method. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or can be a key, trackball or touchpad arranged on the shell of the computer device. It can also be an external keyboard, touchpad or mouse, etc.

[0078] Those skilled in the art can understand, Figure 2The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0079] In an embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above method embodiments when executing the computer program.

[0080] In an embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program implementing the steps in the above method embodiments when executed by a processor.

[0081] In an embodiment, a computer program product is provided, including a computer program, and the computer program implementing the steps in the above method embodiments when executed by a processor.

[0082] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0083] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0084] The above-described embodiments are merely illustrative of several embodiments of the present application, which are described in more detail and specifically, but should not be understood as limiting the scope of the present application. It can be understood that the present application is described through some embodiments, and those skilled in the art know that various changes or equivalent replacements can be made to the features and embodiments without departing from the spirit and scope of the present application. In addition, under the guidance of the present application, the features and embodiments can be modified to adapt to specific conditions and materials without departing from the spirit and scope of the present application. Therefore, the present application is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of the present application are within the scope of the present application.

Claims

1. A concentrator network switching decision system based on communication quality evaluation, characterized in that: include: The communication quality assessment module is used to collect multi-dimensional indicators of the physical layer, network layer, and application layer, and generate a communication quality score Q through preprocessing, dynamic weight allocation, and a comprehensive scoring model; Security protection module, including radio frequency fingerprint recognition unit and digital signature verification unit, used to verify the legitimacy of candidate network signal sources; A handover decision module, which generates a handover decision based on the communication quality score Q and the signal source legitimacy verification result in combination with a dual threshold mechanism; The privacy protection module implements privacy protection of handover decision data through dynamic differential privacy processing and federated learning models; The communication quality assessment module interacts bidirectionally with the security protection module, the switching decision module, and the privacy protection module to form a cross-layer collaborative security protection and decision-making closed loop.

2. The system according to claim 1, wherein: The multi-dimensional indicators include: Physical layer indicators: received signal strength, signal-to-noise ratio, bit error rate, multipath fading coefficient; Network layer indicators: transmission delay, throughput, packet loss rate, link stability; Application layer indicators: service type priority, data integrity requirements, and remaining battery capacity.

3. The system according to claim 1, wherein: The communication quality assessment module includes: The pre-processing unit uses Kalman filtering to denoise the time-varying indicators and extract a 50-dimensional feature vector containing time series characteristics; Dynamic weight allocation unit, which constructs the indicator weight matrix based on the improved analytic hierarchy process and entropy weight method. When the indicator information entropy changes by more than 20%, the weight is recalculated. The comprehensive scoring unit uses a lightweight random forest model to output the communication quality score Q and sets the score correction factor based on the service type.

4. The system according to claim 1, wherein: The security protection module includes: The RF fingerprint recognition unit uses a CNN-LSTM network to extract hardware-specific features such as the signal's carrier frequency offset and phase noise, and performs cosine similarity matching with a database of legitimate AP features. The digital signature verification unit verifies the ECC signature in the beacon frame. The signature content contains a timestamp and a random number to prevent replay attacks.

5. The system according to claim 1, wherein: The switching decision module includes: The dynamic trust calculation unit generates the signal source trust based on the formula TrustScore = 0.35 × RF fingerprint matching + 0.3 × signature verification result + 0.25 × (Q / 100) + 0.1 × historical handover success rate; Dual-threshold trigger unit, setting safety threshold and quality threshold to jointly trigger switching; The pre-switching unit predicts the quality trend of the next 200ms based on the ARIMA model. When the current network Q is predicted to be less than 50 points and the candidate network Q is greater than 80 points, the pre-switching unit starts 50ms in advance.

6. The system according to claim 1, wherein: The privacy protection module includes: Dynamic differential privacy unit, which injects Laplace noise into spatiotemporal trajectory data, and the noise standard deviation is adaptively adjusted according to the movement speed; The federated learning unit adopts a layered architecture to aggregate local model gradient parameters at edge nodes, performs global aggregation through secure multi-party computing in the cloud, and adds Gaussian noise when updating gradients.

7. A concentrator network switching decision method based on communication quality evaluation, characterized in that: Including steps: S1. Multi-dimensional indicator collection and preprocessing: Real-time collection of physical, network, and application layer indicators. Kalman filtering and denoising are used to extract the time series feature vector containing the first four cycles of data. S2. Comprehensive communication quality assessment: Dynamically assign indicator weights using a modified analytic hierarchy process and entropy weighting method, generate a communication quality score (Q) using a random forest model, and adjust the score based on service type. S3. Signal source legitimacy verification: Simultaneously performs RF fingerprint matching and digital signature verification; S4. Cross-layer collaborative handover decision-making: Based on the communication quality score Q and signal source verification results, a dual-threshold mechanism (security threshold and quality threshold) is used to generate handover decisions, triggering pre-handover resource reservation or maintaining the current connection. S5. Privacy-preserving processing: Perform dynamic differential privacy processing on collected data and achieve secure aggregation of model parameters through federated learning.

8. The method according to claim 7, characterized in that The dynamic allocation of indicator weights includes: A three-layer indicator system is constructed, with the weights of the physical layer, network layer, and application layer being 0.4, 0.35, and 0.25, respectively. When the information entropy change of any indicator layer exceeds 20%, the subjective weight is corrected based on the entropy weight method.

9. The method according to claim 7, characterized in that The dual threshold mechanism is specifically: The handover decision is triggered only when the candidate network satisfies the following conditions simultaneously: the radio frequency fingerprint matching degree is greater than 0.85, the digital signature verification is passed, and the candidate network Q is greater than the current network Q+15 points and Q>70 points.

10. The method according to claim 8, characterized in that The dynamic differential privacy processing includes: For high-speed scenarios with moving speeds > 30km / h, the standard deviation of Laplace noise is increased to twice that of low-speed scenarios; privacy protection parameters are dynamically adjusted through an exponential mechanism.

Images