Wireless household appliance intelligent identification and analysis method and system based on multi-protocol fusion

Abstract

The invention belongs to the technical field of wireless household electrical appliance intelligence, and discloses a wireless household electrical appliance intelligent identification and analysis method and system based on multi-protocol fusion, which can automatically extract new equipment protocol features through centralized middleware scheduling and adaptive learning mechanism of a protocol adaptation engine unit in combination with dynamic model training of an AI identification center unit. Generating a temporary adaptive interface and updating the protocol template; the dynamic protocol library supports OTA updating, a novel protocol and unknown equipment can be adapted without manual configuration, and the defects that the number of traditional system equipment is increased, and protocol type expansion is blocked are overcome; the intelligent optimization module dynamically adjusts the bandwidth priority and the data reporting frequency based on reinforcement learning and behavior prediction, guarantees smooth communication of key equipment in a high-concurrency scene, reduces standby power consumption of idle equipment, solves the problems of high-concurrency delay and high energy consumption of a traditional system, and improves the system reliability. And double improvement of identification reliability and system energy efficiency is realized.

Description

Technical Field

[0001] This invention belongs to the field of wireless home appliance intelligent technology, specifically a method and system for wireless home appliance intelligent identification and analysis based on multi-protocol fusion. Background Technology

[0002] With the rapid popularization of the smart home industry, the communication protocols of wireless home appliances have become significantly more diverse, with mainstream protocols including Wi-Fi, Bluetooth, ZigBee, and LoRa. Because manufacturers have not established a unified protocol adaptation standard, devices using different protocols cannot directly interact with each other or control commands. This not only reduces the user experience but also leads to low management efficiency of the home smart system.

[0003] In the prior art, patent CN202410901728.3 discloses a smart home control system with multi-protocol access, which completes protocol conversion through a home host. However, it adopts a distributed protocol management architecture and does not have a unified resource scheduling layer. When the number of devices increases or new protocols are added, it is easy to cause management chaos and conversion conflicts. Moreover, the protocol conversion depends on a predefined device data table and cannot independently adapt to new private protocols and unknown devices, thus limiting its scalability.

[0004] Patent CN202410382813.4 proposes a smart home control system with protocol fusion. It collects data through sensor modules and processes the interaction through protocol fusion modules. However, device identification relies on a static sensor address mapping table and uses the traditional rule of matching "address-device type" one by one. The identification accuracy of second-hand devices or devices from niche brands without address marking is low. Adding new devices requires manual updating of the mapping table and cannot support plug-and-play.

[0005] Furthermore, existing systems require manual reconfiguration of parameters during protocol updates and iterations, making adaptive learning impossible; device identification relies solely on a single identifier or rule matching without incorporating multi-dimensional feature analysis, resulting in insufficient robustness in identification. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for intelligent identification and analysis of wireless home appliances based on multi-protocol fusion, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a wireless home appliance intelligent identification and analysis system based on multi-protocol fusion, the system comprising:

[0008] Protocol adaptation engine unit: includes dynamic protocol library, middleware scheduling and adaptive learning submodule. The adaptive learning submodule uses reinforcement learning to optimize conversion rules; the middleware scheduling uses LSTM to predict communication changes and dynamically allocate resources; the improved algorithm realizes plug-and-play and establishes temporary interfaces for unknown devices.

[0009] AI recognition central unit: includes rule matching, PCA-Adaboost model and meta-learning sub-module. The meta-learning sub-module uses MAML algorithm to identify new devices, receives data from protocol adaptation engine unit to train and update rules; expands the dimension of feature fingerprint library, performs self-supervised pre-training; optimizes process by dynamic weight adjustment, and initiates manual assistance when confidence is low;

[0010] Dynamic parsing unit: Includes incremental parsing mechanism, context awareness and instruction conversion module. Incremental parsing splits instructions by function and caches results to reduce overhead; context awareness expands information dimensions and dynamically adjusts strategies; instruction conversion module uses semantic optimization mechanism to adjust rules and supports custom templates.

[0011] The intelligent optimization module integrates bandwidth allocation and energy management into a multi-objective reinforcement learning system, which autonomously learns optimization strategies; it builds predictive models based on historical data, reserves resources, and adjusts reporting strategies; the device collaborative analysis submodule analyzes and identifies correlation patterns, and uses multi-objective algorithms to balance performance and energy consumption.

[0012] Security control gateway unit: integrates blockchain authentication, zero-knowledge proof and anomaly detection model, blockchain stores device identification and logs, and zero-knowledge proof realizes security authentication; anomaly detection integrates static rules and dynamic models, establishes a threat level mechanism, and synchronizes detection results to the intelligent optimization module to temporarily adjust device priority.

[0013] Preferably, the protocol adaptation engine unit is used for:

[0014] (1) It includes a dynamic protocol library, a middleware scheduling module and an adaptive learning submodule. The three work together to achieve intelligent adaptation of multi-protocol devices. The adaptive learning submodule analyzes the communication behavior of devices through reinforcement learning, establishes a protocol conversion effect evaluation system (including conversion power and delay) and dynamically adjusts the mapping rules. It introduces multimodal protocol recognition technology, integrates deep learning and traditional pattern matching to improve the recognition capability of non-standard protocols, and the generated protocol feature data is synchronized to the feature fingerprint library of the AI ​​recognition central unit in real time to provide basic support for device recognition.

[0015] Formula for evaluating the overall effectiveness of protocol conversion:

[0016] E=α·S-β·D

[0017] In the formula: E represents the overall performance index of protocol conversion, which is used to measure the overall effect of protocol conversion; the higher the value, the better the conversion effect. When E is lower than the threshold, the submodule will automatically trigger the mapping rule adjustment process.

[0018] α represents the conversion power weighting coefficient, with a value range of (0, 1]. It is dynamically set by the system according to the equipment type (such as security equipment that requires a higher success rate) to reflect the importance of conversion power in overall performance. For example, α can be set to 0.7 for key equipment.

[0019] S represents the protocol conversion power, which is the ratio of the number of instructions that successfully complete the protocol conversion within a unit of time (e.g., 1 minute) to the total number of instructions. It is statistically analyzed in real time by the self-adaptive learning submodule to reflect the effectiveness of the current mapping rule.

[0020] β represents the conversion delay weighting coefficient, with a value range of (0, 1]. It is positively correlated with the device's sensitivity to delay. For example, β for real-time control equipment can be set to 0.6 to reflect the degree of impact of delay on overall performance.

[0021] D represents the average protocol conversion latency, which is the average time (in milliseconds) for all instructions to complete the protocol conversion per unit of time. It is collected by the middleware scheduling module and reflects the efficiency of the current resource allocation and conversion algorithm.

[0022] (2) The middleware scheduling module collects Chinese architecture and has protocol prediction function: it establishes a behavior model by mining historical communication data of devices through LSTM model, and loads the protocol conversion module in advance when it detects changes in communication mode; it also has load prediction function, dynamically allocates resources according to device access trend to alleviate high concurrency bottleneck, and synchronizes the predicted changes in communication demand to the bandwidth allocation sub-module of the intelligent optimization module; it achieves plug-and-play through improved protocol feature extraction algorithm (combined with attention mechanism), and the middleware creates temporary interfaces for devices with unknown protocols. The adaptive learning sub-module transmits feature data to the AI ​​recognition central unit, and the new template is synchronized to the dynamic protocol library and dynamic parsing unit via OTA.

[0023] In existing technologies, the distributed protocol management architecture disclosed in CN202410901728.3 lacks a unified resource scheduling layer, which easily leads to management chaos and conversion conflicts when the number of devices increases or new protocols are added. This invention adopts a centralized middleware scheduling architecture, coupled with an LSTM prediction model (the model is trained based on historical device communication data to mine device communication patterns and establish a behavioral model). When a change in communication mode is detected, the corresponding protocol conversion module is pre-loaded. Combined with dynamic resource allocation (prioritizing devices sensitive to transmission delay, such as security equipment having higher priority than ordinary lighting fixtures), the conflict problems of the distributed architecture can be effectively avoided. Simultaneously, the reward function of the reinforcement learning in the adaptive learning submodule is constructed according to conversion power (weight 0.6) and conversion delay (weight 0.4) (referencing the core dimensions of the protocol conversion effect evaluation system), ensuring that the rule optimization direction aligns with actual application needs.

[0024] Preferably, the AI ​​recognition central unit is used for:

[0025] (1) It includes a rule matching module, a PCA-Adaboost model and a meta-learning sub-module. The multi-module collaboration achieves high-precision and fast recognition. The meta-learning sub-module adopts the MAML algorithm to design a fast-adaptive architecture. It can identify new devices with only a small number of labeled samples, solving the dependence of traditional machine learning on large-scale labeled data. It works in conjunction with the protocol adaptation engine unit. After receiving unknown device feature data for training, it will synchronize the updated recognition rules to the feature fingerprint library. The feature fingerprint library expands the device feature dimension, including device operating status and environmental association features. It establishes a feature similarity calculation model, introduces self-supervised learning to pre-train with unlabeled data, and updates the results to the dynamic parsing unit.

[0026] (2) The unit design dynamic weight adjustment mechanism optimizes the dual recognition process: the rule matching and PCA-Adaboost model weights are adjusted according to the equipment type. For equipment of known brands, the rule matching weights are increased to accelerate recognition, and for equipment of unknown brands, the model decision weights are increased; a new recognition confidence assessment module is added, and the manual auxiliary recognition interface is automatically started when the confidence is low; the recognition results (including equipment type, manufacturer information, and function list) are transmitted to the dynamic parsing unit to provide a basis for it to call the corresponding protocol parsing rules;

[0027] In existing technologies, CN202410382813.4 uses a static mapping of "address-device type" to identify devices. This results in low accuracy for secondhand devices without address labeling or devices from niche brands, and requires manual updating of the mapping table for new devices. The meta-learning submodule of this invention employs the MAML algorithm (inner loop learning rate 0.01, outer loop learning rate 0.001), requiring only 5-10 labeled samples (containing basic device protocol features and typical functional instruction codes) to train the novel device identification model, eliminating the need for large-scale labeled data. Furthermore, the feature fingerprint database is expanded to include device operating features (such as the power fluctuation range of air conditioners and the brightness adjustment step size of lamps) and environmental association features (such as the ambient temperature range when the device starts up). Combined with self-supervised pre-training (using unlabeled device data, 50 pre-training rounds), this further improves the robustness of identification and solves the problems of low accuracy and lack of plug-and-play support for unlabeled devices caused by static mapping.

[0028] Formula for calculating device identification confidence:

[0029] C=ω1·C R +ω2·C M

[0030] ω1+ω2=1

[0031] In the formula: C represents the device recognition confidence level, which is used to judge the reliability of the recognition result. The value range is [0, 1]. When C≥0.8, the recognition result is directly adopted. When C<0.5, the manual auxiliary recognition interface is activated.

[0032] ω1 represents the rule matching weight, with a value range of (0, 1). It is set by a dynamic weight adjustment mechanism based on whether the device brand is known. For devices with known brands, ω1 can be set to 0.8 (preferred rule matching), and for devices with unknown brands, it can be set to 0.3.

[0033] C R C represents the rule matching confidence score, indicating the degree of matching between device features and known rules in the feature fingerprint database. Its value ranges from [0, 1]. If the features are a perfect match, then C is the confidence score. R =1, the lower the matching degree, the smaller the value;

[0034] ω2 represents the machine learning weight, with a value range of (0,1). It is complementary to ω1. For unknown brand devices, ω2 can be set to 0.7 (prioritizing the PCA-Adaboost model results), and for known brands, it can be set to 0.2.

[0035] C M The confidence score represents the PCA-Adaboost model's recognition confidence level, indicating the reliability of the model's judgment on device type. The value range is [0, 1], and it is output by the model through multiple rounds of classification training, reflecting the accuracy of feature dimensionality reduction and classification.

[0036] Preferably, the dynamic parsing unit is used for:

[0037] (1) It includes an incremental parsing mechanism, a context-aware module, and an instruction conversion module to achieve efficient parsing and cross-protocol conversion of device instructions. The incremental parsing mechanism draws on the line-level parsing concept of VSCode, dividing instructions into independent parsing units according to functional units and establishing a caching mechanism. When the instruction part changes, only the changed part is re-parsed. An incremental update data structure is designed to realize the dynamic loading of parsing rules. The rules come from the device recognition results of the AI ​​recognition hub unit and the new protocol templates of the protocol adaptation engine unit. The parsed standardized dataset is synchronized to the intelligent optimization module for formulating optimization strategies.

[0038] (2) The context awareness module expands the information dimensions. In addition to network load and energy consumption level, it adds device correlation information (device collaborative analysis data from the intelligent optimization module) and user behavior habits (user preference data from the intelligent optimization module). It establishes a dynamic adjustment model for context weights and optimizes the parsing strategy according to the importance of the device (from the AI ​​recognition central unit) and the real-time status. For example, it raises the priority when the security camera is abnormal, and the result is synchronized to the intelligent optimization module. The instruction conversion module is based on the unified abstract model and introduces a semantic mapping dynamic optimization mechanism. It supports the management of custom instruction templates. The private protocol extension definition pushed by the manufacturer's OTA is synchronized to the relevant library after verification, and the conversion result is synchronized to the security control gateway unit.

[0039] In existing technologies, conventional smart home optimization systems often employ single-objective optimization (such as optimizing only response latency), failing to balance performance and energy consumption. This invention's multi-objective reinforcement learning system uses the DDPG algorithm (each Actor network and Critic network contains two hidden layers, with 256 neurons per layer). The reward function is set according to network throughput (weight 0.4), response latency (weight 0.3), and device energy consumption (weight 0.3) (weight allocation is based on the priority of device usage in a home scenario; security and temperature control devices have higher requirements for throughput and latency). Simultaneously, a behavior prediction model is established based on historical device operating data (collecting data from the past 7 days, with daily sampling points set at fixed intervals). The bandwidth reservation is set at 1.2 times the prediction requirement, and the data reporting strategy adjusts the reporting frequency during off-peak periods (such as 2-4 AM, referencing users' daily usage patterns), achieving a balance between performance and energy consumption and overcoming the limitations of single-objective optimization.

[0040] Preferably, the intelligent optimization module is used for:

[0041] (1) Includes sub-modules for bandwidth allocation, energy management and device collaborative analysis, and realizes dynamic resource optimization through reinforcement learning and prediction models; the bandwidth allocation and energy management sub-modules are integrated into a multi-objective reinforcement learning system, defining a reward function with network throughput, response latency and energy consumption as the core, and using the DDPG algorithm to learn optimization strategies autonomously; a device priority dynamic evaluation model is established, which evaluates according to device type (from the AI ​​recognition central unit), usage scenario (from the dynamic parsing unit) and user habits (from the parsing data), and the generated optimization strategy is synchronized to the dynamic parsing unit to adjust the parsing priority, and the middleware scheduling module of the protocol adaptation engine unit to allocate resources;

[0042] Formula for calculating device bandwidth allocation:

[0043]

[0044] In the formula: B i This represents the bandwidth allocation for the i-th device (unit: Mbps), and is the core output of the bandwidth allocation submodule, ensuring that high-priority, high-demand devices receive sufficient bandwidth resources.

[0045] P i This represents the priority coefficient of the i-th device, with a value range of [1, 5]. It is determined by the device priority dynamic evaluation model based on the device type (e.g., security camera P). i =5) Use case settings; the higher the value, the higher the priority.

[0046] R iThe predicted communication demand of the i-th device (in Mbps) is calculated by the behavioral prediction model based on the device's historical operating data (such as the average bandwidth usage in the previous hour) and reflects the device's bandwidth demand over a future period.

[0047] n represents the total number of devices currently connected to the system. It is calculated in real time by the protocol adaptation engine and is dynamically updated as devices connect / disconnect. It is one of the basic parameters for bandwidth allocation.

[0048] B total This represents the total available bandwidth of the system (unit: Mbps), obtained in real time by the bandwidth allocation submodule based on network environment monitoring. For example, if the total bandwidth of a home network is 100 Mbps, then B... total =100;

[0049] (2) Based on the historical operating data of the device in the dynamic parsing unit, a prediction model is built to accurately predict changes in communication demand: the bandwidth allocation submodule reserves resources to avoid congestion based on the prediction results, the energy consumption management submodule adjusts the data reporting strategy according to the low usage period of the device to reduce ineffective energy consumption, and the prediction results are synchronized to the protocol prediction function module of the protocol adaptation engine unit; the device collaborative analysis module identifies the associated usage patterns by analyzing the device linkage data to achieve group-level resource optimization; a multi-objective optimization algorithm is designed to ensure the performance of key devices (based on the priority of the central unit identified by AI) while balancing system energy consumption, and the optimization results are synchronized to the dynamic parsing unit to adjust the group device parsing strategy.

[0050] Preferably, the security control gateway unit is used for:

[0051] (1) Integrate blockchain authentication mechanism, zero-knowledge proof technology and anomaly detection model to build a full-process security protection system; the blockchain authentication mechanism establishes a device identity authentication system based on consortium blockchain, and stores the device's unique identifier (from the protocol adaptation engine unit) and certificate information on the blockchain to ensure immutability; when the device accesses, its identity is verified by the blockchain, and the result is fed back to the protocol adaptation engine unit to decide whether to allow access. It also realizes on-chain storage of device operation logs (including adaptation, identification, parsing and optimization links) to ensure traceability and solve the risks of traditional centralized authentication; zero-knowledge proof technology (such as zk-SNARKs) is applied to authentication, and the device can be verified without disclosing sensitive information; establish an attribute-based encryption scheme, realize fine-grained permission management based on the device information of the AI ​​recognition central unit, and synchronize the encryption rules to the instruction conversion module of the dynamic parsing unit;

[0052] (2) The anomaly detection model integrates static rules and dynamic behavior analysis: static rules cover known attack patterns, while the dynamic behavior model learns the normal communication characteristics of devices (from historical data of the dynamic parsing unit) through an autoencoder to identify unknown anomalies; a security threat level assessment mechanism is established to automatically adjust the protection strategy according to the severity of the threat, from interception to isolation of suspicious devices; the threat detection results are synchronized to the intelligent optimization module in real time for temporary adjustment of device priority and resource allocation; and the security of device access and command transmission is guaranteed throughout the process.

[0053] In existing technologies, conventional smart home systems mostly employ centralized device authentication, which poses a single point of failure risk, and anomaly detection only covers known attack patterns. This invention's blockchain authentication mechanism is built on a consortium blockchain (containing three types of nodes: home gateway, cloud management, and manufacturer authentication, with node consensus using the PBFT algorithm). Unique device identifiers (composed of a manufacturer-assigned device code and a manufacturing timestamp) and certificate information are stored on the blockchain and cannot be tampered with. Simultaneously, the anomaly detection model's static rules cover known attack patterns (such as illegal IP command sending and excessive control requests, referencing common IoT attack characteristics), while the dynamic model uses an autoencoder (input being device communication characteristics over the past 5 minutes, including command sending frequency and data frame length, with a reconstruction error threshold set to 0.3), capable of identifying unknown anomalies and solving the challenges of centralized authentication risks and unknown threat detection.

[0054] Security threat level assessment formula:

[0055] T = γ·T S +(1-γ)·T D

[0056] In the formula: T represents the security threat level, with a value range of 1.5. It is the basis for adjusting the security protection strategy. Levels 1-2 only record logs, while levels 4-5 trigger device isolation.

[0057] γ represents the static rule matching weight, with a value range of (0, 1). The default value is 0.4, and it can be increased to 0.6 when the known attack patterns occur frequently, reflecting the fundamental role of static rules in threat assessment.

[0058] T S The static rule threat level is represented by a value in the range [1, 5], which is determined by the anomaly detection model by comparing device instructions with a known attack rule base. A perfect match for a high-risk rule results in T. S =5, T when there is no match S =1;

[0059] T D The dynamic behavior threat level is represented by a value in the range [1, 5]. It is calculated by an autoencoder based on the deviation between the device's communication characteristics and the normal behavior model. The larger the deviation, the higher the threat level. DThe higher the value, such as when the command sending frequency is abnormal, T D =4.

[0060] This invention also provides a method for intelligent identification and analysis of wireless home appliances based on multi-protocol fusion. Based on the above system, the specific steps of this method are as follows:

[0061] S1. Dynamic Protocol Adaptation and Device Discovery: When a new device is input, the protocol adaptation engine unit detects and pre-identifies the protocol and pre-loads the template, and performs initial verification on the blockchain of the security gateway; for known devices, the template is adapted and the features are transmitted to the AI ​​recognition central unit; for unknown devices, a temporary interface is established and the features are transmitted to the AI ​​recognition central unit to train the template; the new template is OTA synchronized to the dynamic parsing unit, the edge-cloud collaborative adaptation is started, and the protocol template and feature data are output.

[0062] S2. Multi-dimensional intelligent recognition and data analysis: Following the device features and protocol templates in step S1, the AI ​​recognition central unit initiates three-level recognition and outputs the results and confidence levels to the dynamic analysis unit; the dynamic analysis unit calls the rules according to the confidence level, and uses context assistance for low confidence levels to incrementally analyze and generate a standardized dataset; the data is synchronized to the intelligent optimization module, and error cases are fed back to the AI ​​recognition central unit, which outputs the dataset and analysis strategy;

[0063] S3. Adaptive Optimization and Security Control: Following the dataset and parsing strategy in step S2, the intelligent optimization module generates an optimization strategy and adjusts the parsing priority; the security control gateway unit encrypts instructions and detects anomalies, and the encrypted instructions are transmitted to the protocol adaptation engine unit, triggering alarms when anomalies occur; after execution, the device provides data feedback, the threat is resolved and the normal strategy is restored, the user preference module adjusts the weights, and outputs the optimization strategy and security instructions.

[0064] Preferably, the specific steps of dynamic protocol adaptation and device discovery in step S1 are as follows:

[0065] S11. When a new device is input, the protocol adaptation engine unit discovers the device through "active detection + passive listening" and simultaneously initiates protocol feature pre-identification and identity verification: it quickly determines the device's protocol category based on lightweight feature matching and preloads relevant protocol templates; the security control gateway unit performs preliminary verification of the new device based on blockchain authentication technology, allowing only verified devices to conduct basic communication; for devices with successful protocol matching, the protocol adaptation engine unit calls dynamic protocol library templates to complete the adaptation and synchronizes the device's basic features to the feature fingerprint database of the AI ​​recognition hub unit;

[0066] S12. For devices with unknown protocols, the middleware scheduling module creates a temporary adaptation interface to ensure basic communication. The adaptive learning submodule extracts the feature data of the unknown protocol and uploads it to the AI ​​recognition central unit in real time to start the model training and new protocol template generation process. After the new protocol template is updated to the dynamic protocol library via OTA, it is automatically synchronized to the dynamic parsing unit to support subsequent instruction parsing. The edge-cloud collaborative adaptation mechanism is started simultaneously, and nodes are allocated according to the complexity of protocol conversion: simple conversions are completed at the edge nodes to reduce latency, and complex conversions are processed by the cloud to ensure accuracy. The processing results are fed back to the middleware scheduling module of the protocol adaptation engine unit to optimize subsequent resource allocation and output the protocol template and device feature data.

[0067] Preferably, the specific steps of multi-dimensional intelligent recognition and data parsing in step S2 are as follows:

[0068] S21. Following the device characteristics and protocol templates from step S1, the AI ​​recognition central unit initiates a "three-level recognition architecture": Level 1 uses the rule matching module to quickly filter known devices by calling historical rules from the feature fingerprint database; Level 2 uses the meta-learning submodule to initially classify and narrow down the scope of suspected new devices; Level 3 uses the PCA-Adaboost model to complete accurate recognition, outputting results including device type, manufacturer information, function list, and confidence score, which are synchronized to the dynamic parsing unit to provide a basis for calling parsing rules.

[0069] S22. The dynamic parsing unit calls the protocol parsing rules based on the confidence score: high-confidence results directly apply the matching rules, while low-confidence results initiate context-assisted parsing (combined with data from the intelligent optimization module) to improve accuracy; at the same time, an incremental parsing mechanism is initiated to re-parse the device-reported instructions, cache the results, generate a standardized dataset, and output the dataset and parsing strategy; the standardized dataset is synchronized to the intelligent optimization module for formulating optimization strategies; parsing error cases are automatically fed back to the AI ​​recognition central unit to update the feature fingerprint database and recognition model; parsing rule update requirements are synchronized to the protocol adaptation engine unit to ensure matching with the new protocol template.

[0070] Preferably, the specific steps of adaptive optimization and safety control in step S3 are as follows:

[0071] S31. Following the standardized dataset and parsing strategy from step S2, the intelligent optimization module receives the data and generates an adaptive optimization strategy through a multi-objective reinforcement learning system: the bandwidth allocation submodule reserves resources based on the device priority and behavior prediction model results of the AI ​​recognition central unit; the energy consumption management submodule adjusts the data reporting strategy according to the low usage period of the equipment; the device collaborative analysis module formulates a group optimization plan based on the linkage data; the optimization strategy is synchronized to the dynamic parsing unit to adjust the device instruction parsing priority and ensure that critical equipment has priority in parsing resources.

[0072] S32. The security control gateway intervenes throughout the process: it performs AES encryption and anomaly detection (based on a dynamic behavior model) on optimization commands, filters malicious commands, and transmits legitimate commands to the protocol adaptation engine unit to adjust the protocol conversion rules; device operation logs are stored on the blockchain, and abnormal commands trigger alarms and are synchronized to the intelligent optimization module, temporarily adjusting the priority of abnormal devices; after the optimization strategy is executed, the device operation data is fed back to the intelligent optimization module through the dynamic parsing unit; after the security threat is eliminated, the security control gateway notifies the module to restore the normal strategy, and the user preference learning module adjusts the optimization target weights based on the feedback data.

[0073] The beneficial effects of this invention are as follows:

[0074] 1. This invention utilizes a centralized middleware scheduling and adaptive learning mechanism in the protocol adaptation engine unit, combined with dynamic model training in the AI ​​recognition hub unit, to automatically extract protocol features of new devices, generate temporary adaptation interfaces, and update protocol templates. The dynamic protocol library supports OTA updates, enabling adaptation to new protocols and unknown devices without manual configuration. This solves the problem of traditional systems experiencing lag as the number of devices and protocol types increase, achieving seamless access and efficient collaboration for various wireless home appliances and significantly improving the protocol compatibility and scalability of smart home systems.

[0075] 2. The AI ​​recognition central unit of this invention improves the device recognition accuracy to a higher level through a three-level recognition architecture of "rule matching + meta-learning + multi-dimensional feature fingerprint library" and avoids misidentification of unlabeled devices. The intelligent optimization module dynamically adjusts bandwidth priority and data reporting frequency based on reinforcement learning and behavior prediction, ensuring smooth communication of key devices in high-concurrency scenarios, while reducing the standby power consumption of idle devices. This solves the pain points of "high concurrency latency and high energy consumption" in traditional systems and achieves a dual improvement in recognition reliability and system energy efficiency.

[0076] 3. The security control gateway unit of this invention integrates blockchain authentication, zero-knowledge proof, and anomaly detection to ensure that device identity is tamper-proof, sensitive information is not leaked, and malicious commands can be intercepted. At the same time, the adaptive learning capabilities of each unit reduce manual intervention, and protocol updates and rule optimizations are completed automatically. This forms a security closed loop of "access-identification-control," solving the security vulnerabilities of traditional systems, and significantly shortening operation and maintenance time. It also reduces the security risks and management costs of smart home systems and improves user security and convenience. Attached Figure Description

[0077] Figure 1 This is a flowchart of the wireless home appliance intelligent identification and analysis system based on multi-protocol fusion of the present invention;

[0078] Figure 2This is a flowchart of the wireless home appliance intelligent identification and analysis method based on multi-protocol fusion according to the present invention. Detailed Implementation

[0079] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0080] like Figures 1 to 2 As shown, this embodiment of the invention provides a method and system for intelligent identification and analysis of wireless home appliances based on multi-protocol fusion. The system includes:

[0081] Protocol adaptation engine unit: includes dynamic protocol library, middleware scheduling and adaptive learning submodule. The adaptive learning submodule uses reinforcement learning to optimize conversion rules; the middleware scheduling uses LSTM to predict communication changes and dynamically allocate resources; the improved algorithm realizes plug-and-play and establishes temporary interfaces for unknown devices.

[0082] AI recognition central unit: includes rule matching, PCA-Adaboost model and meta-learning sub-module. The meta-learning sub-module uses MAML algorithm to identify new devices, receives data from protocol adaptation engine unit to train and update rules; expands the dimension of feature fingerprint library, performs self-supervised pre-training; optimizes process by dynamic weight adjustment, and initiates manual assistance when confidence is low;

[0083] Dynamic parsing unit: Includes incremental parsing mechanism, context awareness and instruction conversion module. Incremental parsing splits instructions by function and caches results to reduce overhead; context awareness expands information dimensions and dynamically adjusts strategies; instruction conversion module uses semantic optimization mechanism to adjust rules and supports custom templates.

[0084] The intelligent optimization module integrates bandwidth allocation and energy management into a multi-objective reinforcement learning system, which autonomously learns optimization strategies; it builds predictive models based on historical data, reserves resources, and adjusts reporting strategies; the device collaborative analysis submodule analyzes and identifies correlation patterns, and uses multi-objective algorithms to balance performance and energy consumption.

[0085] Security control gateway unit: integrates blockchain authentication, zero-knowledge proof and anomaly detection model, blockchain stores device identification and logs, and zero-knowledge proof realizes security authentication; anomaly detection integrates static rules and dynamic models, establishes a threat level mechanism, and synchronizes detection results to the intelligent optimization module to temporarily adjust device priority.

[0086] The protocol adaptation engine unit is used for:

[0087] (1) It includes a dynamic protocol library, a middleware scheduling module and an adaptive learning submodule. The three work together to achieve intelligent adaptation of multi-protocol devices. The adaptive learning submodule analyzes the communication behavior of devices through reinforcement learning, establishes a protocol conversion effect evaluation system (including conversion power and delay) and dynamically adjusts the mapping rules. It introduces multimodal protocol recognition technology, integrates deep learning and traditional pattern matching to improve the recognition capability of non-standard protocols, and the generated protocol feature data is synchronized to the feature fingerprint library of the AI ​​recognition central unit in real time to provide basic support for device recognition.

[0088] The specific rules for traditional pattern matching in multimodal protocol identification include, for example, matching the fixed field "0x0801" in the frame header of the Wi-Fi protocol; the data transmission of the temporary interface for unknown devices adopts JSON format, with fields including "Device ID-Protocol Feature-Transmission Time", which clarifies the data interaction format and ensures that the data is standardized and consistent when the interface is used;

[0089] The dynamic protocol library update mechanism: After the new protocol template is synchronized via OTA, the dynamic protocol library first performs version verification on the template (the verification field includes "template version number - compatible device type - update time"). If the verification passes, the old template is replaced and the old template is backed up for 3 days. If the verification fails, OTA retransmission is triggered, and the number of retransmissions shall not exceed 3. The protocol library version management and exception handling logic are clearly defined to avoid template update chaos.

[0090] The dynamic protocol library's storage structure: The dynamic protocol library stores templates in a three-level classification of "protocol type-manufacturer-device model", such as "Wi-Fi-Xiaomi-Air Conditioner KFR-35GW". Each template folder contains "protocol rule file, feature description file, and compatible device list". When calling a template, the protocol type is matched first, and then the manufacturer + model is used for precise positioning. The storage and calling logic is clear, avoiding low template retrieval efficiency or incorrect positioning.

[0091] (2) The middleware scheduling module collects Chinese architecture and has protocol prediction function: it establishes a behavior model by mining historical communication data of devices through LSTM model, and loads the protocol conversion module in advance when it detects changes in communication mode; it also has load prediction function, dynamically allocates resources according to device access trend to alleviate high concurrency bottleneck, and synchronizes the predicted changes in communication demand to the bandwidth allocation submodule of the intelligent optimization module; it achieves plug-and-play through improved protocol feature extraction algorithm (combined with attention mechanism), and creates temporary interfaces for devices with unknown protocols. The adaptive learning submodule transmits feature data to the AI ​​recognition central unit, and the new template is synchronized to the dynamic protocol library and dynamic parsing unit via OTA;

[0092] The input data for the LSTM model scheduled by the middleware consists of the device communication frequency and data volume over the past 30 minutes, ensuring that the source of the model training data is clear. The reward function of the reinforcement learning in the adaptive learning submodule is constructed with conversion power (weight 0.6) and conversion delay (weight 0.4), making the reward calculation logic feasible and avoiding the inability to reproduce due to ambiguous parameters.

[0093] The AI ​​recognition central unit is used for:

[0094] (1) It includes a rule matching module, a PCA-Adaboost model and a meta-learning sub-module. The multi-module collaboration achieves high-precision and fast recognition. The meta-learning sub-module adopts the MAML algorithm to design a fast-adaptive architecture. It can identify new devices with only a small number of labeled samples, solving the dependence of traditional machine learning on large-scale labeled data. It works in conjunction with the protocol adaptation engine unit. After receiving unknown device feature data for training, it will synchronize the updated recognition rules to the feature fingerprint library. The feature fingerprint library expands the device feature dimension, including device operating status and environmental association features. It establishes a feature similarity calculation model, introduces self-supervised learning to pre-train with unlabeled data, and updates the results to the dynamic parsing unit.

[0095] The training sample format for new devices in the meta-learning submodule is as follows: Samples must include "basic device characteristics (model, protocol type), operating characteristics (power, communication frequency), and example of functional instructions (such as 'temperature adjustment to 26℃' instruction code)". The data size of each sample should be controlled within 512 bytes. During training, the training set and validation set should be divided in an "8:2" ratio. The training is considered successful when the accuracy of the validation set is ≥90%. Clear sample specifications and training acceptance criteria are established to ensure that the training of the new device recognition model has a basis.

[0096] The specific features of the feature fingerprint database include, for example, the device operating status including air conditioning cooling / heating mode, and the environmental association features including temperature (unit °C) and humidity (unit %). It is clear that the feature dimension of the PCA-Adaboost model after dimensionality reduction is 32 and the number of weak classifiers is 50, so that the model parameters are clear and the model training process can be reproduced.

[0097] The inner loop learning rate of the MAML algorithm is set to 0.01 and the outer loop learning rate is set to 0.001, thus clarifying the algorithm training parameters. After the manual assistance recognition interface is started, it supports administrators to upload device model and manufacturer information via the web interface, and the interface response time is ≤3 seconds. The operation process is refined to ensure that the manual assistance link is executable.

[0098] The fingerprint database update frequency and conflict resolution mechanism: The fingerprint database is automatically updated once every 24 hours. When adding a device feature, if the similarity with the old feature is ≥0.9 (calculated by cosine similarity), it is judged as a conflict. The latest feature is retained first and the old feature is marked as "to be deleted". If no associated device is used after 7 days, it will be deleted. The fingerprint database maintenance rules are clear to avoid feature redundancy or conflict.

[0099] (2) The unit design dynamic weight adjustment mechanism optimizes the dual recognition process: the rule matching and PCA-Adaboost model weights are adjusted according to the equipment type. For equipment of known brands, the rule matching weight is increased to accelerate recognition, and the model decision weight is increased for equipment of unknown brands. A new recognition confidence assessment module is added. When the confidence is low, the manual auxiliary recognition interface is automatically started. The recognition results (including equipment type, manufacturer information, and function list) are transmitted to the dynamic parsing unit to provide a basis for it to call the corresponding protocol parsing rules.

[0100] The dynamic parsing unit is used for:

[0101] (1) It includes an incremental parsing mechanism, a context-aware module, and an instruction conversion module to achieve efficient parsing and cross-protocol conversion of device instructions. The incremental parsing mechanism draws on the line-level parsing concept of VSCode, dividing instructions into independent parsing units according to functional units and establishing a caching mechanism. When the instruction part changes, only the changed part is re-parsed. An incremental update data structure is designed to realize the dynamic loading of parsing rules. The rules come from the device recognition results of the AI ​​recognition hub unit and the new protocol templates of the protocol adaptation engine unit. The parsed standardized dataset is synchronized to the intelligent optimization module for formulating optimization strategies.

[0102] The incremental parsing cache is stored in Redis, with the cache key format being "Device ID-Unit ID-Timestamp" and the expiration time set to 15 minutes. When the instruction part changes, only the changed unit is re-parsed. For example, if the temperature value changes, only the "Temperature Adjustment" unit is parsed. This clarifies the caching logic and the scope of re-parsing, avoiding cache management chaos.

[0103] (2) The context awareness module expands the information dimensions. In addition to network load and energy consumption level, it adds device correlation information (device collaborative analysis data from the intelligent optimization module) and user behavior habits (user preference data from the intelligent optimization module). It establishes a dynamic adjustment model for context weights and optimizes the parsing strategy according to the importance of the device (from the AI ​​recognition central unit) and the real-time status. For example, it raises the priority when the security camera is abnormal, and the result is synchronized to the intelligent optimization module. The instruction conversion module is based on the unified abstract model and introduces a semantic mapping dynamic optimization mechanism. It supports the management of custom instruction templates. The private protocol extension definition pushed by the manufacturer's OTA is synchronized to the relevant library after verification, and the conversion result is synchronized to the security control gateway unit.

[0104] In context awareness, device association information comes from device linkage data from the intelligent optimization module, such as the linkage relationship between "air conditioner and temperature and humidity sensor"; the semantic mapping rule example of the instruction conversion module, such as mapping the air conditioner "Temp+1" instruction to the standard instruction "SET_TEMP_INC", provides a specific basis for the semantic optimization mechanism and facilitates rule adjustment;

[0105] The weight adjustment logic of the context awareness module is as follows: The weight calculation formula for device importance (e.g., security camera is level 5, ordinary lamp is level 2) and real-time status (abnormal status is 1, normal status is 0.5) is "context weight = device importance × 0.6 + real-time status × 0.4"; when the weight is ≥3, the parsing priority is increased, and when it is <2, the priority is decreased. The rules for weight calculation and priority association are clearly defined to avoid the lack of quantitative standards for parsing strategy adjustment.

[0106] Custom instruction template creation and verification process: When creating a template, the administrator needs to fill in "template name - compatible device type - instruction fields (including field meaning, data type, and value range)"; during verification, simulate parsing 10 test instructions, and if the pass rate is ≥95%, the verification is successful; otherwise, the template needs to be modified, and the template creation specifications and verification standards should be clearly defined to ensure that the custom template can be used normally.

[0107] The intelligent optimization module is used for:

[0108] (1) Includes sub-modules for bandwidth allocation, energy management and device collaborative analysis, and realizes dynamic resource optimization through reinforcement learning and prediction models; the bandwidth allocation and energy management sub-modules are integrated into a multi-objective reinforcement learning system, defining a reward function with network throughput, response latency and energy consumption as the core, and using the DDPG algorithm to learn optimization strategies autonomously; a device priority dynamic evaluation model is established, which evaluates according to device type (from the AI ​​recognition central unit), usage scenario (from the dynamic parsing unit) and user habits (from the parsing data), and the generated optimization strategy is synchronized to the dynamic parsing unit to adjust the parsing priority, and the middleware scheduling module of the protocol adaptation engine unit to allocate resources;

[0109] The DDPG algorithm parameters for the multi-objective reinforcement learning system are as follows: the Actor network contains 2 hidden layers (256 neurons per layer), and the Critic network contains 2 hidden layers (256 neurons per layer); the experience replay pool size is set to 10000, and 64 experience samples are collected for training each time. The network structure and training parameters are clearly defined to ensure that the algorithm can be reproduced during training and to avoid training the model without a basis.

[0110] The iteration termination condition for the multi-objective reinforcement learning system is as follows: When training the DDPG algorithm, the iteration is terminated when the reward function value fluctuates by ≤5% for 10 consecutive iterations and the satisfaction rate of "latency ≤100ms and energy consumption ≤1.5kWh" on the validation set is ≥95%. The maximum number of iterations is set to 200 rounds to avoid infinite training loops or termination before reaching the optimal effect.

[0111] The weighting of device priority evaluation indicators is as follows: device type accounts for 0.5, usage scenario accounts for 0.3, and user habits account for 0.2. For example, the priority of security cameras is set to 5. The group optimization of device collaboration analysis is based on the frequency of device linkage. For example, if the linkage frequency of "light-human sensor" is high, resources will be allocated first, so that priority evaluation and group optimization are systematic.

[0112] (2. Based on the device's historical operating data from the dynamic parsing unit, a prediction model is built to accurately predict changes in communication demand: the bandwidth allocation submodule reserves resources to avoid congestion based on the prediction results, the energy consumption management submodule adjusts the data reporting strategy according to the device's off-peak usage period to reduce ineffective energy consumption, and the prediction results are synchronized to the protocol prediction function module of the protocol adaptation engine unit; the device collaborative analysis module identifies associated usage patterns by analyzing device linkage data to achieve group-level resource optimization; a multi-objective optimization algorithm is designed to ensure the performance of key devices (based on AI to identify the priority of the central unit) while balancing system energy consumption, and the optimization results are synchronized to the dynamic parsing unit to adjust the group device parsing strategy;

[0113] The behavioral prediction model is trained using equipment operation data from the past 7 days, with a deviation of ≤10% in predicting communication demand; bandwidth allocation reserves resources at 1.2 times the predicted demand; the data reporting strategy reduces the frequency to once every 5 minutes during off-peak equipment usage periods (2-4 AM); and the sources of predicted data and optimization execution details are clearly defined.

[0114] The security control gateway unit is used for:

[0115] (1) Integrate blockchain authentication mechanism, zero-knowledge proof technology and anomaly detection model to build a full-process security protection system; the blockchain authentication mechanism establishes a device identity authentication system based on consortium blockchain, and stores the device's unique identifier (from the protocol adaptation engine unit) and certificate information on the blockchain to ensure immutability; when the device accesses, its identity is verified by the blockchain, and the result is fed back to the protocol adaptation engine unit to decide whether to allow access. It also realizes on-chain storage of device operation logs (including adaptation, identification, parsing and optimization links) to ensure traceability and solve the risks of traditional centralized authentication; zero-knowledge proof technology (such as zk-SNARKs) is applied to authentication, and the device can be verified without disclosing sensitive information; establish an attribute-based encryption scheme, realize fine-grained permission management based on the device information of the AI ​​recognition central unit, and synchronize the encryption rules to the instruction conversion module of the dynamic parsing unit;

[0116] The specific implementation process of zero-knowledge proof: When a device is connected, it submits a "device identifier hash-permission attribute proof". The verification node verifies it through the zk-SNARKs algorithm. First, it verifies that the hash is consistent with the chain, and then it verifies whether the attribute proof meets the permission policy (such as ordinary users can only verify basic permissions). The verification takes ≤1 second. The authentication steps and time limits are clearly defined to ensure that the security authentication can be implemented.

[0117] The attribute-based encryption scheme key distribution mechanism is as follows: the key is generated by the key management module of the security control gateway unit and distributed through the "secure channel when the device first connects (based on TLS1.3)"; the key validity period is set to 90 days, and a new key is automatically pushed 7 days before the expiration date. After receiving the key, the device needs to provide confirmation information, which clarifies the key generation, distribution and update process and avoids key leakage or device decryption.

[0118] (2) The anomaly detection model integrates static rules and dynamic behavior analysis: static rules cover known attack patterns, while the dynamic behavior model learns the normal communication characteristics of devices (from historical data of the dynamic parsing unit) through an autoencoder to identify unknown anomalies; a security threat level assessment mechanism is established to automatically adjust the protection strategy according to the severity of the threat, from interception to isolation of suspicious devices; the threat detection results are synchronized to the intelligent optimization module in real time for temporary adjustment of device priority and resource allocation; and the security of device access and command transmission is guaranteed throughout the process.

[0119] The corresponding measures for security threat levels are as follows: Levels 1-2 only log, Level 3 intercept commands, and Levels 4-5 isolate devices; the device identification field of the blockchain storage is in the format of "DEV-2024-001", and operation logs are retained for 30 days. The security policy execution method and data storage specifications are clearly defined to ensure that security controls can be implemented.

[0120] Examples of static rules for anomaly detection include prohibiting devices with the IP address "192.168.1.100" from sending commands; the dynamic model uses an autoencoder, with the input being the communication characteristics of devices over the past 5 minutes, and anomalies are determined when the reconstruction error is greater than 0.3. Clearly define the detection rules and model judgment criteria to avoid ambiguity in anomaly detection logic.

[0121] This invention also provides a method for intelligent identification and analysis of wireless home appliances based on multi-protocol fusion. Based on the above system, the specific steps of this method are as follows:

[0122] S1. Dynamic Protocol Adaptation and Device Discovery: When a new device is input, the protocol adaptation engine unit detects and pre-identifies the protocol and pre-loads the template, and performs initial verification on the blockchain of the security gateway; for known devices, the template is adapted and the features are transmitted to the AI ​​recognition central unit; for unknown devices, a temporary interface is established and the features are transmitted to the AI ​​recognition central unit to train the template; the new template is OTA synchronized to the dynamic parsing unit, the edge-cloud collaborative adaptation is started, and the protocol template and feature data are output.

[0123] S2. Multi-dimensional intelligent recognition and data analysis: Following the device features and protocol templates in step S1, the AI ​​recognition central unit initiates three-level recognition and outputs the results and confidence levels to the dynamic analysis unit; the dynamic analysis unit calls the rules according to the confidence level, and uses context assistance for low confidence levels to incrementally analyze and generate a standardized dataset; the data is synchronized to the intelligent optimization module, and error cases are fed back to the AI ​​recognition central unit, which outputs the dataset and analysis strategy;

[0124] S3. Adaptive Optimization and Security Control: Following the dataset and parsing strategy in step S2, the intelligent optimization module generates an optimization strategy and adjusts the parsing priority; the security control gateway unit encrypts instructions and detects anomalies, and the encrypted instructions are transmitted to the protocol adaptation engine unit, triggering alarms when anomalies occur; after execution, the device provides data feedback, the threat is resolved and the normal strategy is restored, the user preference module adjusts the weights, and outputs the optimization strategy and security instructions.

[0125] The specific steps of dynamic protocol adaptation and device discovery in step S1 are as follows:

[0126] S11. When a new device is input, the protocol adaptation engine unit discovers the device through "active detection + passive listening" and simultaneously initiates protocol feature pre-identification and identity verification: it quickly determines the device's protocol category based on lightweight feature matching and preloads relevant protocol templates; the security control gateway unit performs preliminary verification of the new device based on blockchain authentication technology, allowing only verified devices to conduct basic communication; for devices with successful protocol matching, the protocol adaptation engine unit calls dynamic protocol library templates to complete the adaptation and synchronizes the device's basic features to the feature fingerprint database of the AI ​​recognition hub unit;

[0127] The protocol adaptation engine unit actively probes every 5 seconds, and the timeout for passively listening to and capturing broadcast packets is 10 seconds. The initial blockchain verification requires the approval of two or more nodes: the home gateway and the cloud management. The probe frequency and the number of verification nodes are clearly defined to avoid unclear execution standards in the probe and verification process.

[0128] S12. For devices with unknown protocols, the middleware scheduling module creates a temporary adaptation interface to ensure basic communication. The adaptive learning submodule extracts the feature data of the unknown protocol and uploads it to the AI ​​recognition central unit in real time to start the model training and new protocol template generation process. After the new protocol template is updated to the dynamic protocol library via OTA, it is automatically synchronized to the dynamic parsing unit to support subsequent instruction parsing. The edge-cloud collaborative adaptation mechanism is started simultaneously, and nodes are allocated according to the complexity of protocol conversion: simple conversions are completed at the edge nodes to reduce latency, and complex conversions are processed by the cloud to ensure accuracy. The processing results are fed back to the middleware scheduling module of the protocol adaptation engine unit to optimize subsequent resource allocation and output the protocol template and device feature data.

[0129] Edge-cloud collaborative adaptation node communication protocol: The edge and cloud use the MQTT protocol to transmit transformation task data, with the topic set to "edge-cloud / task" and the data format including "task ID-transformation complexity-device characteristics"; after the cloud completes the processing, it will return the result through the same topic, with the communication timeout set to 3 seconds, to clarify the node data interaction specifications and avoid abnormal data transmission during collaborative adaptation.

[0130] The temporary interface of an unknown device will be closed under the following conditions: when the new protocol template is synchronized to the dynamic parsing unit via OTA and the device completes three normal protocol conversions using the new template, the temporary interface will be automatically closed; if the normal conversion is not completed within one hour, an interface fault alarm will be triggered, requiring manual investigation to clarify the interface lifecycle and fault handling, and to avoid the temporary interface occupying resources for a long time.

[0131] OTA updates use a differential update method, with the update package size controlled within 100KB; in edge-cloud collaborative adaptation, edge nodes handle tasks with ≤3 steps of conversion, while the cloud handles tasks with >3 steps, refining the update method and task division criteria to ensure that collaborative adaptation is operable.

[0132] The specific steps of multi-dimensional intelligent recognition and data parsing in step S2 are as follows:

[0133] S21. Following the device characteristics and protocol templates from step S1, the AI ​​recognition central unit initiates a "three-level recognition architecture": Level 1 uses the rule matching module to quickly filter known devices by calling historical rules from the feature fingerprint database; Level 2 uses the meta-learning submodule to initially classify and narrow down the scope of suspected new devices; Level 3 uses the PCA-Adaboost model to complete accurate recognition, outputting results including device type, manufacturer information, function list, and confidence score, which are synchronized to the dynamic parsing unit to provide a basis for calling parsing rules.

[0134] The threshold for matching the first-level rules of the AI ​​identification central unit is 100%, meaning that when the device characteristics are completely consistent with the rules in the fingerprint database, it is determined to be a known device; the standardized dataset adopts JSON format, and the fields include "device ID-instruction type-parsing result-timestamp", which clearly defines the identification criteria and data format and avoids confusion between identification and data storage;

[0135] S22. The dynamic parsing unit invokes protocol parsing rules based on the confidence score: high-confidence results directly apply matching rules, while low-confidence results initiate context-assisted parsing (combined with data from the intelligent optimization module) to improve accuracy; simultaneously, an incremental parsing mechanism is activated to re-parse the device-reported instructions, cache the results, generate a standardized dataset, and output the dataset and parsing strategy; the standardized dataset is synchronized to the intelligent optimization module for formulating optimization strategies; parsing error cases are automatically fed back to the AI ​​recognition central unit to update the feature fingerprint database and recognition model; parsing rule update requirements are synchronized to the protocol adaptation engine unit to ensure matching with the new protocol template;

[0136] The feedback process for parsing error cases is as follows: After parsing error cases are fed back to the AI ​​recognition central unit, the central unit automatically extracts error features (such as "instruction field missing - parsing failure"), adds 5 new similar labeled samples for model fine-tuning, and regenerates parsing rules after fine-tuning and synchronizes them to the dynamic parsing unit. This clarifies the error iteration mechanism and ensures continuous optimization of parsing accuracy.

[0137] When dynamically parsing unit context aids, the example of user behavior habits combined is "the user adjusts the air conditioner temperature at 7 pm every day"; the cache validity period for incremental parsing is set to 5 minutes, and it is automatically refreshed after the timeout, clearly defining the context aid basis and cache management rules to ensure that the parsing strategy is executable;

[0138] For low confidence levels (0.5 ≤ C < 0.8), the weight allocation of contextual aids is as follows: the weight of device correlation information is set to 0.6, and the weight of user behavior habits is set to 0.4. For example, "the linkage between the air conditioner and the temperature and humidity sensor (correlation weight 0.6) + the user adjusts the temperature at 19:00 (habit weight 0.4)", when the comprehensive score is ≥ 0.7, the resolution confidence is increased to above 0.8. The logic of auxiliary information weight and confidence improvement is clearly defined to avoid inaccurate resolution results with low confidence.

[0139] The specific steps of adaptive optimization and safety control in step S3 are as follows:

[0140] S31. Following the standardized dataset and parsing strategy from step S2, the intelligent optimization module receives the data and generates an adaptive optimization strategy through a multi-objective reinforcement learning system: the bandwidth allocation submodule reserves resources based on the device priority and behavior prediction model results of the AI ​​recognition central unit; the energy consumption management submodule adjusts the data reporting strategy according to the low usage period of the equipment; the device collaborative analysis module formulates a group optimization plan based on the linkage data; the optimization strategy is synchronized to the dynamic parsing unit to adjust the device instruction parsing priority and ensure that critical equipment has priority in parsing resources.

[0141] The intelligent optimization module reserves bandwidth resources at 1.2 times the predicted demand, and adjusts the data reporting interval to 10 minutes during off-peak hours (2-4 am); the security control gateway unit uses the AES-128 algorithm for encryption commands, and the key is dynamically allocated by the gateway, clearly defining optimization parameters and encryption details to ensure that optimization and encryption are feasible.

[0142] S32. The security control gateway intervenes throughout the process: it performs AES encryption and anomaly detection (based on a dynamic behavior model) on optimization commands, filters malicious commands, and transmits legitimate commands to the protocol adaptation engine unit to adjust the protocol conversion rules; device operation logs are stored on the blockchain, and abnormal commands trigger alarms and are synchronized to the intelligent optimization module, temporarily adjusting the priority of abnormal devices; after the optimization strategy is executed, the device operation data is fed back to the intelligent optimization module through the dynamic parsing unit; after the security threat is eliminated, the security control gateway notifies the module to restore the normal strategy, and the user preference learning module adjusts the optimization target weights based on the feedback data;

[0143] Anomalies from the security control gateway unit are synchronized to the user's app, allowing users to be informed promptly. When adjusting weights in the user preference module, the optimization weight for frequently used devices is set to 0.6, clearly defining the alarm reception method and weight adjustment standard, thus providing a concrete basis for security alarms and preference optimization.

[0144] The weight adjustment cycle of the user preference module: The module adjusts the weights every day at 2:00 AM based on user behavior data from the past 7 days (such as device usage frequency and operating habits). The optimization weight of frequently used devices (usage frequency ≥ 5 times / day) is increased by 0.1, with a maximum of 0.8. The timing and magnitude of weight updates are clearly defined to ensure that preference optimizations are aligned with user behavior.

[0145] The verification steps for restoring the normal policy after threat resolution are as follows: After the threat is resolved, the security control gateway unit first sends a "test command (such as 'query current status')" to the device. If the device responds normally and there are no new anomalies within 5 minutes, the normal policy is restored. If the response is abnormal, the temporary policy is maintained and an alarm is issued. This clarifies the restoration verification process and avoids security risks caused by restoring the normal policy before the threat is completely resolved.

[0146] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0147] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A wireless home appliance intelligent identification and analysis system based on multi-protocol fusion, characterized in that: The system includes: Protocol adaptation engine unit: includes dynamic protocol library, middleware scheduling and adaptive learning submodule. The adaptive learning submodule uses reinforcement learning to optimize conversion rules; the middleware scheduling uses LSTM to predict communication changes and dynamically allocate resources; the improved algorithm realizes plug-and-play and establishes temporary interfaces for unknown devices. AI recognition central unit: includes rule matching, PCA-Adaboost model and meta-learning sub-module. The meta-learning sub-module uses MAML algorithm to identify new devices, receives data from protocol adaptation engine unit to train and update rules; expands the dimension of feature fingerprint library, performs self-supervised pre-training; optimizes process by dynamic weight adjustment, and initiates manual assistance when confidence is low; Dynamic parsing unit: Includes incremental parsing mechanism, context awareness and instruction conversion module. Incremental parsing splits instructions by function and caches results to reduce overhead; context awareness expands information dimensions and dynamically adjusts strategies; instruction conversion module uses semantic optimization mechanism to adjust rules and supports custom templates. The intelligent optimization module integrates bandwidth allocation and energy management into a multi-objective reinforcement learning system, which autonomously learns optimization strategies; it builds predictive models based on historical data, reserves resources, and adjusts reporting strategies; the device collaborative analysis submodule analyzes and identifies correlation patterns, and uses multi-objective algorithms to balance performance and energy consumption. Security control gateway unit: integrates blockchain authentication, zero-knowledge proof and anomaly detection model, blockchain stores device identification and logs, and zero-knowledge proof realizes security authentication; anomaly detection integrates static rules and dynamic models, establishes a threat level mechanism, and synchronizes detection results to the intelligent optimization module to temporarily adjust device priority.

2. The wireless home appliance intelligent identification and analysis system based on multi-protocol fusion according to claim 1, characterized in that: The protocol adaptation engine unit is used for: (1) It includes a dynamic protocol library, a middleware scheduling module and an adaptive learning submodule. The adaptive learning submodule analyzes the device communication behavior through reinforcement learning, establishes a protocol conversion effect evaluation system and dynamically adjusts the mapping rules; introduces multimodal protocol recognition technology, integrates deep learning and traditional pattern matching to improve the non-standard protocol recognition capability, and the generated protocol feature data is synchronized to the feature fingerprint library of the AI ​​recognition central unit in real time. (2) The middleware scheduling module collects Chinese architecture and has protocol prediction function: it establishes a behavior model by mining historical communication data of devices through LSTM model, and loads the protocol conversion module in advance when it detects changes in communication mode; it also has load prediction function, dynamically allocates resources according to device access trend, and synchronizes the predicted changes in communication demand to the bandwidth allocation sub-module of intelligent optimization module; it achieves plug-and-play through improved protocol feature extraction algorithm, and creates temporary interfaces for devices with unknown protocols through middleware, and the adaptive learning sub-module transmits feature data to the AI ​​recognition central unit, and the new template is synchronized to the dynamic protocol library and dynamic parsing unit via OTA.

3. The wireless home appliance intelligent identification and analysis system based on multi-protocol fusion according to claim 2, characterized in that: The AI ​​recognition central unit is used for: (1) It includes a rule matching module, a PCA-Adaboost model and a meta-learning sub-module. The meta-learning sub-module uses the MAML algorithm to identify new devices. It works in conjunction with the protocol adaptation engine unit to receive unknown device feature data for training and then synchronizes the updated identification rules to the feature fingerprint database. The feature fingerprint database expands the dimensions of device features, establishes a feature similarity calculation model, introduces self-supervised learning to pre-train with unlabeled data, and updates the results to the dynamic parsing unit. (2) Design a dynamic weight adjustment mechanism to optimize the dual recognition process: Adjust the rule matching and PCA-Adaboost model weights according to the equipment type. Extract the rule matching weights for known brand equipment to accelerate recognition, and increase the model decision weights for unknown brand equipment; add a recognition confidence assessment module, and automatically start the manual auxiliary recognition interface when the confidence is low; the recognition result is transmitted to the dynamic parsing unit.

4. The wireless home appliance intelligent identification and analysis system based on multi-protocol fusion according to claim 3, characterized in that: The dynamic parsing unit is used for: (1) Includes an incremental parsing mechanism, a context-aware module, and an instruction conversion module. The incremental parsing mechanism draws on the line-level parsing concept of VSCode, dividing instructions into independent parsing units according to functional units and establishing a caching mechanism. When the instruction part changes, only the changed part is re-parsed. An incremental update data structure is designed to realize the dynamic loading of parsing rules. The rules come from the device recognition results of the AI ​​recognition hub unit and the new protocol template of the protocol adaptation engine unit. The parsed standardized dataset is synchronized to the intelligent optimization module. (2) The context-aware module expands the information dimensions, adding device association information and user behavior habits in addition to network load and energy consumption level; Establish a dynamic adjustment model for context weights, optimize the parsing strategy based on device importance and real-time status, and synchronize the results to the intelligent optimization module; The instruction conversion module is based on a unified abstraction model, introduces a dynamic semantic mapping optimization mechanism, supports custom instruction template management, and synchronizes the private protocol extension definitions pushed by the manufacturer via OTA to the relevant library after verification. The conversion results are synchronized to the security control gateway unit.

5. The wireless home appliance intelligent identification and analysis system based on multi-protocol fusion according to claim 4, characterized in that: The intelligent optimization module is used for: (1) It includes sub-modules for bandwidth allocation, energy management and device collaborative analysis. The bandwidth allocation and energy management sub-modules are integrated into a multi-objective reinforcement learning system. A reward function is defined with network throughput, response latency and energy consumption as the core. The DDPG algorithm is used to learn optimization strategies autonomously. Establish a dynamic evaluation model for device priority, evaluate based on device type, usage scenario, and user habits, and synchronize the generated optimization strategy to the middleware scheduling module of the dynamic parsing unit and protocol adaptation engine unit; (2) Based on the historical operating data of the device in the dynamic parsing unit, a behavior prediction model is established to predict changes in communication demand: the bandwidth allocation submodule reserves resources according to the prediction results, the energy consumption management submodule adjusts the data reporting strategy according to the low usage period of the device, and the prediction results are synchronized to the protocol prediction function module of the protocol adaptation engine unit. The equipment collaborative analysis module identifies associated usage patterns by analyzing equipment linkage data; it designs multi-objective optimization algorithms to ensure the performance of key equipment while balancing system energy consumption, and the optimization results are synchronized to the dynamic analysis unit.

6. The wireless home appliance intelligent identification and analysis system based on multi-protocol fusion according to claim 5, characterized in that: The security control gateway unit is used for: (1) Integrating blockchain authentication mechanism, zero-knowledge proof technology and anomaly detection model, the blockchain authentication mechanism builds a device identity authentication system based on consortium blockchain, and stores the device's unique identifier and certificate information on the blockchain; when the device accesses, its identity is verified by the blockchain, and the result is fed back to the protocol adaptation engine unit to decide whether to allow access. It also realizes on-chain storage of device operation logs. Zero-knowledge proof technology is applied to authentication, and the device can be verified without disclosing sensitive information; an attribute-based encryption scheme is established, and the encryption rules are synchronized to the dynamic parsing unit. (2) The anomaly detection model integrates static rules and dynamic behavior analysis: static rules cover known attack patterns, while the dynamic behavior model learns the normal communication characteristics of the device through an autoencoder to identify unknown anomalies. Establish a security threat level assessment mechanism to automatically adjust protection strategies based on the severity of the threat; threat detection results are synchronized to the intelligent optimization module in real time for temporary adjustments to device priorities and resource allocation.

7. A method for intelligent identification and analysis of wireless home appliances based on multi-protocol fusion, based on the system described in claim 6, characterized in that: The specific steps of this method are as follows: S1. Dynamic Protocol Adaptation and Device Discovery: When a new device is input, the protocol adaptation engine unit detects and pre-identifies the protocol and pre-loads the template, and performs initial verification on the blockchain of the security gateway; known devices are adapted using templates and transmit features to the AI ​​recognition hub unit; unknown devices establish a temporary interface and transmit features to the AI ​​recognition hub unit to train templates; the new template is OTA synchronized to the dynamic parsing unit, the edge-cloud collaborative adaptation is started, and the protocol template and feature data are output. S2. Multi-dimensional intelligent recognition and data parsing: Following the device features and protocol templates in step S1, the AI ​​recognition central unit initiates three-level recognition and outputs the results and confidence levels to the dynamic parsing unit; the dynamic parsing unit calls the rules according to the confidence level, and uses context assistance for low confidence levels to incrementally parse and generate a standardized dataset; Data is synchronized to the intelligent optimization module, and error cases are fed back to the AI ​​recognition central unit, which outputs the dataset and analysis strategy. S3. Adaptive Optimization and Security Control: Following the dataset and parsing strategy in step S2, the intelligent optimization module generates optimization strategies and adjusts parsing priorities; The security control gateway unit encrypts commands and detects anomalies. The encrypted commands are transmitted to the protocol adaptation engine unit, and anomalies trigger alarms. After execution, the device provides data feedback, the threat is resolved and the normal policy is restored, the user preference module adjusts the weights, and the optimized policy and security instructions are output.

8. The method for intelligent identification and analysis of wireless home appliances based on multi-protocol fusion according to claim 7, characterized in that: The specific steps of dynamic protocol adaptation and device discovery in step S1 are as follows: S11. When a new access device is input, the protocol adaptation engine unit discovers the device through active detection and passive listening, and simultaneously starts protocol feature pre-identification and identity verification: the device protocol category is determined based on lightweight feature matching, and relevant protocol templates are preloaded. The security control gateway unit performs preliminary verification of new devices based on blockchain authentication technology, allowing verified devices to conduct basic communication; for devices with successful protocol matching, the protocol adaptation engine unit calls the dynamic protocol library template to complete the adaptation and synchronizes the device's basic characteristics to the AI ​​recognition central unit; S12. For devices with unknown protocols, the middleware scheduling module creates a temporary adaptation interface to ensure basic communication. The adaptive learning submodule extracts the feature data of the unknown protocol and uploads it to the AI ​​recognition central unit in real time to start the model training and new protocol template generation process. After the new protocol template is updated to the dynamic protocol library via OTA, it is automatically synchronized to the dynamic parsing unit. The edge-cloud collaborative adaptation mechanism is started simultaneously, and nodes are allocated according to the complexity of protocol conversion: simple conversions are completed at the edge nodes to reduce latency, and complex conversions are processed by the cloud to ensure accuracy. The processing results are fed back to the middleware scheduling module of the protocol adaptation engine unit, and the protocol template and device feature data are output.

9. The method for intelligent identification and analysis of wireless home appliances based on multi-protocol fusion according to claim 8, characterized in that: The specific steps of multi-dimensional intelligent recognition and data parsing in step S2 are as follows: S21. Following the device characteristics and protocol templates from step S1, the AI ​​recognition central unit initiates a three-level recognition architecture: Level 1 uses the rule matching module to call historical rules from the feature fingerprint database to filter known devices; Level 2 uses the meta-learning submodule to initially classify and narrow down the scope of suspected new devices; Level 3 completes the recognition through the PCA-Adaboost model, outputting results including device type, manufacturer information, function list, and confidence score, which are then synchronized to the dynamic parsing unit. S22. The dynamic parsing unit calls the protocol parsing rules according to the confidence score: the matching rules are directly applied to high confidence results, and context-assisted parsing is started for low confidence results; the incremental parsing mechanism is started to re-parse the device-reported instructions, generate a standardized dataset after caching the results, and output the dataset and parsing strategy. The standardized dataset is synchronized to the intelligent optimization module; Error cases are automatically fed back to the AI ​​recognition central unit, and the requirements for updating the parsing rules are synchronized to the protocol adaptation engine unit.

10. A method for intelligent identification and analysis of wireless home appliances based on multi-protocol fusion according to claim 9, characterized in that: The specific steps of adaptive optimization and security control in step S3 are as follows: S31. Following the standardized dataset and parsing strategy of step S2, the intelligent optimization module receives the data and generates an adaptive optimization strategy through a multi-objective reinforcement learning system: the bandwidth allocation submodule reserves resources by combining the device priority of the AI ​​recognition central unit with the behavior prediction model results. The energy management submodule adjusts the data reporting strategy based on the off-peak usage period of the equipment; The equipment collaborative analysis module formulates group optimization plans based on linkage data; the optimization strategy is synchronized to the dynamic parsing unit to adjust the priority of equipment command parsing; S32. The security control gateway unit intervenes throughout the process: it performs AES encryption and anomaly detection on optimization commands, filters malicious commands, and then transmits legitimate commands to the protocol adaptation engine unit to adjust the protocol conversion rules; the device operation log is stored on the blockchain, and abnormal commands trigger alarms and are synchronized to the intelligent optimization module to temporarily adjust the priority of abnormal devices. After the optimization strategy is executed, the equipment operation data is fed back to the intelligent optimization module via the dynamic analysis unit; after the security threat is eliminated, the normal strategy is restored, and the user preference learning module adjusts the target weights based on the feedback data.

Images