Cross-regional laboratory management method and system based on protocol adaptive analysis

Abstract

The invention relates to the technical field of industrial automation and artificial intelligence crossing, in particular to a cross-regional laboratory management method and system based on protocol adaptive analysis. The method comprises the following steps: monitoring equipment communication data flow through edge nodes deployed in each region, automatically identifying and analyzing a protocol by using a protocol adaptive analysis module based on a deep learning model, generating standardized data, and uploading the standardized data to a cloud platform; the cloud platform analyzes data through a global cooperative control engine based on deep reinforcement learning, dynamically generates a cross-regional optimization control strategy, and issues and executes the cross-regional optimization control strategy; meanwhile, the platform performs equipment anomaly detection and predictive maintenance based on a time sequence prediction model, and triggers a protective control strategy when an anomaly is detected. According to the invention, plug and play of equipment, global intelligent scheduling and optimization of resources and predictive maintenance of faults are realized, and automation and intelligence levels of cross-regional laboratory management and system reliability are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial automation and artificial intelligence, in particular to a cross-regional laboratory management method and system based on protocol adaptive parsing. BACKGROUND

[0002] In the automation and informatization management of cross-regional laboratories, the traditional distributed management system usually adopts a two-level architecture of "SCADA unit + center cloud platform". The edge unit is responsible for interfacing various experimental equipment, collecting data and performing basic control; the cloud platform performs data aggregation, storage, display and report generation. However, this traditional solution has several unresolved technical bottlenecks in electrical implementation. First, at the device access level, the protocol adaptation of the edge unit to heterogeneous devices is highly dependent on manual programming. Each time a new model or private protocol device is accessed, the developer needs to manually write and debug the parsing code, resulting in long system integration period, high cost, poor flexibility, and inability to meet the needs of rapid iteration and plug-and-play of devices. Secondly, at the control logic level, the automation strategy executed by the edge unit is mostly a fixed rule set in advance, which cannot be dynamically adjusted and optimized according to complex factors such as global device state, real-time load, energy consumption, etc. The control strategy is rigid and difficult to achieve resource coordination and energy efficiency optimization across regions. Thirdly, at the data analysis level, the functions of the cloud platform mostly stop at data visualization and basic statistics, lacking deep mining and prediction ability for device running state, and unable to realize fault warning and predictive maintenance, making the operation and maintenance work passive. In addition, the security of the entire system relies on traditional firewalls and permission controls, lacking active sensing and intelligent defense capabilities for abnormal operations and potential network attacks. The root cause of these problems lies in the insufficient "intelligence" level of existing systems, which fail to deeply integrate cutting-edge artificial intelligence technology with industrial automation scenarios, resulting in weak self-adaptation, self-optimization and autonomous decision-making capabilities. Therefore, there is an urgent need for a new cross-regional laboratory management solution that can intelligently adapt to devices, dynamically optimize control, and have deep prediction capabilities.

[0003] Therefore, the prior art still needs further development. SUMMARY

[0004] The purpose of the present application is to overcome the above technical deficiencies and provide a cross-regional laboratory management method and system based on protocol adaptive parsing to solve the problems existing in the prior art.

[0005] To achieve the above technical purpose, according to the first aspect of the present application, the present application provides a cross-regional laboratory management method based on protocol adaptive parsing, comprising: S1, the edge node deployed in each laboratory area listens to the communication data stream of the newly accessed device in real time; S2, automatically identifying and analyzing the communication data stream through a protocol adaptive analysis module based on a deep learning model, to generate standardized device data; S3, uploading the standardized device data to a cloud platform; S4, analyzing the standardized device data received from multiple edge nodes through a global collaborative control engine based on deep reinforcement learning of the cloud platform, to dynamically generate an optimized control strategy across regions; S5, issuing the optimized control strategy to the corresponding edge node for execution.

[0006] Specifically, in step S2, the deep learning model is a one-dimensional convolutional neural network or a Transformer sequence model.

[0007] Specifically, in step S2, the specific process of automatic identification and analysis includes extracting the time sequence features and frame structure features of the communication data stream, and outputting the analyzed device parameters and state information.

[0008] Specifically, the protocol adaptive analysis module runs on a field programmable gate array or microprocessor of the edge node to ensure the real-time performance of protocol analysis.

[0009] Specifically, in step S4, the optimization objectives of the global collaborative control engine include at least one of system energy efficiency, device load balancing, and task completion efficiency.

[0010] Specifically, in step S4, the process of dynamically generating an optimized control strategy is based on the continuous interaction between a deep reinforcement learning agent and the system environment, and is obtained through trial and error learning.

[0011] Specifically, the optimized control strategy includes adjusting the operating power of a specific device or migrating computing tasks between devices in different regions.

[0012] Specifically, the method further includes step S6: the cloud platform performs device anomaly detection and predictive maintenance alarm through a time series prediction model based on the standardized device data.

[0013] Specifically, when the time series prediction model detects a device anomaly, the global collaborative control engine generates a corresponding protective control strategy.

[0014] According to a second aspect of the present application, a cross-regional laboratory management system based on protocol adaptive analysis is provided, comprising: Edge nodes deployed in each laboratory region, wherein the edge nodes are provided with a protocol adaptive analysis module for automatically identifying and analyzing the communication data stream of the access device to generate standardized device data; The cloud platform communicates with all the edge nodes. The cloud platform is equipped with a global collaborative control engine, which is used to dynamically generate cross-regional optimized control strategies based on the received standardized device data and distribute them to the corresponding edge nodes. The protocol adaptive parsing module is based on a deep learning model, and the global collaborative control engine is based on a deep reinforcement learning model.

[0015] Beneficial effects: The cross-regional laboratory management method and system based on protocol adaptive parsing and intelligent collaborative control provided by this invention brings multiple significant benefits through the deep integration of artificial intelligence technologies such as deep learning and deep reinforcement learning.

[0016] At the device access and data fusion level, this invention achieves "plug-and-play" connectivity for heterogeneous devices by deploying a protocol adaptive parsing module based on a deep learning model at edge nodes. This module can automatically identify and parse the communication protocols of unknown devices without any manual coding intervention, completely solving the core pain points of inflexible device access and poor compatibility in traditional systems. It greatly reduces the technical threshold and cost of system integration, expansion, and maintenance, providing high-quality, standardized real-time data sources for upper-layer intelligent applications.

[0017] At the level of global optimization and collaborative control, this invention introduces a global collaborative control engine based on deep reinforcement learning into the cloud platform, constructing the system's "intelligent hub." This engine can continuously learn and analyze real-time status information from the entire network, dynamically generating and distributing cross-regional optimization control strategies guided by multiple objectives such as system energy efficiency and load balancing. This enables the system to leap from automation that executes fixed rules to intelligence with autonomous decision-making and continuous optimization capabilities, achieving dynamic scheduling and optimal configuration of cross-laboratory resources, significantly improving overall equipment utilization, reducing energy consumption, and ensuring the execution efficiency of critical tasks.

[0018] In terms of predictive maintenance and system resilience, this invention's innovative predictive maintenance mechanism uses a time-series prediction model to perform in-depth analysis of equipment data, enabling the early detection of potential fault signs and achieving a fundamental shift from "reactive maintenance" to "preemptive warning." Combined with the calculated anomaly confidence level, the system can quantitatively assess anomalies and issue tiered alerts, providing precise data for maintenance decisions. Furthermore, when a high-confidence anomaly is detected, the mechanism proactively triggers a collaborative control engine to generate and execute protective strategies, such as load migration or device isolation, effectively preventing fault escalation and ensuring the high availability and operational continuity of the entire laboratory system.

[0019] In summary, this invention constructs an adaptive, self-optimizing, and highly resilient intelligent management system by organically combining edge-side intelligent sensing, cloud-based intelligent decision-making, and predictive maintenance. It not only achieves breakthrough improvements in the flexibility of device access, the global optimization of resource scheduling, and the proactiveness of operation and maintenance management, but also enhances system security through intelligent closed-loop systems. Ultimately, it provides a solid technical foundation for the refined, intelligent, and unmanned operation of cross-regional laboratories, possessing extremely high practical value and promising prospects for widespread adoption. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the cross-regional laboratory management method based on protocol adaptive parsing provided in a specific embodiment of the present invention; Figure 2 This is a schematic diagram of the system composition of a cross-regional laboratory management system based on protocol adaptive parsing provided in a specific embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments in this application, other similar embodiments obtained by those skilled in the art without creative effort should all fall within the scope of protection of this application. Furthermore, directional terms mentioned in the following embodiments, such as "up," "down," "left," and "right," are only for reference to the directions in the accompanying drawings; therefore, the directional terms used are for illustrative purposes and not for limiting the invention.

[0022] The present invention will be further described below with reference to the accompanying drawings and preferred embodiments.

[0023] Please see Figure 1 This invention provides a cross-regional laboratory management method based on protocol adaptive parsing, comprising: S1. Edge nodes deployed in various laboratory areas monitor the communication data streams of newly connected devices in real time. It should be further explained that this invention constructs a complete intelligent closed loop from edge perception to cloud decision-making. The edge node is an industrial control computer (ICC) deployed in each physical laboratory area, with a CPU clock speed of no less than 2.0 GHz, memory of no less than 8 GB, and equipped with multiple RS-485, RS-232, Ethernet, and GPIO interfaces for connecting various laboratory equipment within the area, such as constant temperature and humidity chambers, spectrum analyzers, programmable power supplies, etc. In step S1, the data acquisition service running on the edge node continuously monitors all communication buses at a sampling frequency of 1000 times / second. When a new device connects and begins sending data, the acquisition service can capture its initial communication message in real time and cache at least the first 1024 bytes of data stream for subsequent analysis.

[0024] S2. The communication data stream is automatically identified and parsed by the protocol adaptive parsing module based on a deep learning model to generate standardized device data. Further explanation is needed regarding step S2, which is the core innovation. Its specific implementation is as follows: The protocol adaptive parsing module is triggered, which normalizes the cached raw byte stream (dividing each byte value by 255, mapping it to the [0,1] interval), and then feeds it into a pre-trained deep learning model for inference. The model outputs two key results: the protocol type probability distribution and the starting position and length of each data field within the message. Based on this output, the parsing module extracts specific device parameters (such as temperature value "25.6"), status information (such as device on / off status "1"), and timestamps according to the identified protocol format specifications, ultimately encapsulating them into a standardized data object conforming to the JSON Schema standard.

[0025] S3. Upload the standardized equipment data to the cloud platform; It should be further noted that in step S3, the data object is published to a specified Topic on the cloud platform via an IP Sec VPN encrypted tunnel established over the Internet using the MQTT protocol. To ensure real-time performance, the default data upload frequency is once per second, but for critical devices, it can be configured for real-time upload (latency <100ms).

[0026] S4. Through the cloud platform's global collaborative control engine based on deep reinforcement learning, standardized device data received from multiple edge nodes is analyzed to dynamically generate cross-regional optimized control strategies. It should be further explained that in step S4, the cloud platform's global collaborative control engine subscribes to these MQTT Topics. It maintains a global system state view and periodically (e.g., every 5 seconds) runs deep reinforcement learning inference to generate optimization strategies.

[0027] S5. The optimized control strategy is sent to the corresponding edge nodes for execution; It should be further explained that step S5 ensures policy execution. The engine encodes the policy into JSON instructions and sends them to the target edge node via MQTT. The control service on the node parses the instructions and drives the actuator (such as setting the power supply output voltage via Modbus TCP protocol) or adjusts the local control logic parameters.

[0028] It is understandable that the beneficial effects of this invention are fundamental, realizing a shift from "humans adapting to machines" to "machines adapting to humans." Edge intelligence enables zero-configuration access for devices, while cloud intelligence achieves globally optimal allocation of resources across regions, significantly reducing system integration, maintenance costs, and energy consumption, while simultaneously improving laboratory utilization and operational reliability.

[0029] Specifically, in step S2, the deep learning model is a one-dimensional convolutional neural network or a Transformer sequence model.

[0030] It should be further explained that the deep learning model used in the protocol adaptive parsing module is its core technology carrier.

[0031] Furthermore, the preferred option is a one-dimensional convolutional neural network (1D-CNN), the specific structure of which is shown below. This structure is the preferred structure after extensive experimental verification, achieving a good balance between accuracy and inference speed: 1. Input layer: Receives a byte sequence of fixed length of 256 (padding with zeros if insufficient, truncating if too long), with an input dimension of (256, 1).

[0032] 2. Convolutional Layer 1: Uses 64 one-dimensional convolutional kernels of size 5 with a stride of 1, employing "same" padding to maintain the output length, followed by a ReLU activation function. This layer is used to extract local patterns (such as fixed frame header features) in byte sequences.

[0033] 3. Max pooling layer 1: The pooling size is 2, the stride is 2, and the sequence length is compressed to 128.

[0034] 4. Convolutional Layer 2: Uses 128 one-dimensional convolutional kernels of size 3 with a stride of 1, "same" padding, followed by a ReLU activation function. This layer is used to combine low-level features to form more complex patterns.

[0035] 5. Max pooling layer 2: The pooling size is 2, the stride is 2, and the sequence length is compressed to 64.

[0036] 6. Flattening layer: Flattens the 3D feature map into a 1D vector.

[0037] 7. Fully connected layer 1: Contains 256 neurons, uses the ReLU activation function, followed by a Dropout layer with a Dropout rate of 0.5 to prevent overfitting.

[0038] 8. Fully connected layer 2 (output layer): The number of neurons is equal to the number of protocol categories to be classified (e.g., 10 categories). The Softmax activation function is used to output the probability of each protocol category.

[0039] Furthermore, the preferred second option is the Transformer encoder model, which is better suited for capturing long-range dependencies. The model consists of an input embedding layer (mapping each byte to a 64-dimensional vector), plus positional encoding, followed by three stacked Transformer encoding layers. Each encoding layer includes an 8-head self-attention mechanism and a feedforward network (128 dimensions). Finally, classification is performed through a global average pooling layer and a fully connected output layer. The rationale for choosing between 1D-CNN and Transformer is that 1D-CNN is computationally efficient, well-suited for extracting local features from byte sequences, and easily hardware-accelerated on FPGAs; while the Transformer model has slightly higher computational cost, its self-attention mechanism better understands the global structure of the message, resulting in higher recognition accuracy for complex protocols. Users can choose based on their priorities regarding real-time performance and accuracy.

[0040] Specifically, in step S2, the automatic identification and parsing process includes: extracting the timing features and frame structure features of the communication data stream, and outputting the parsed device parameters and status information.

[0041] It's important to further explain that automatic recognition and parsing is a sophisticated, multi-step process. First, the deep learning model automatically extracts features through its hierarchical structure. For 1D-CNN, shallow convolutional kernels activate common frame start symbols such as 0xAA and 0x55, while deeper kernels can identify combination patterns like "function code-data length". For Transformer, its self-attention weights clearly show the most critical byte positions for protocol judgment (e.g., the 2nd and 3rd bytes of the message typically carry the address and function code, and have the highest attention weights). During model training, we not only use protocol types as labels but also introduce sequence labeling tasks, labeling each byte of the message (e.g., B-Address, I-Address, B-Data, I-Data, O, etc., using the BIO labeling system). This allows the model to accurately locate each data domain while classifying. Specifically, in terms of model architecture, a Conditional Random Field (CRF) layer or a simple fully connected layer can be connected after the feature extraction layers of the 1D-CNN or Transformer to predict the label for each byte position. During model inference, a single forward propagation outputs both the protocol type and the tag for each byte of the message. The parsing module then easily locates the start and end indices of the "data field" based on the tag sequence output by the CRF layer, thereby accurately extracting the original parameter bytes. These bytes are then converted into physically meaningful values ​​according to the protocol specification (e.g., resolving the two bytes 0x01 and 0x2C to the decimal temperature value 300, representing 30.0°C).

[0042] Understandably, the beneficial effect of the above process is that it achieves end-to-end automated parsing, transforming the traditional work that required a lot of expert experience and manual programming into a data-driven model training process, which greatly improves efficiency and reduces error rate.

[0043] Specifically, the protocol adaptive parsing module runs on the field-programmable gate array or microprocessor of the edge node to ensure the real-time performance of protocol parsing.

[0044] It should be further noted that, in order to meet industrial-grade real-time requirements (the total time for a single protocol identification and parsing process should be less than 10 milliseconds), the protocol adaptive parsing module is preferably implemented using hardware acceleration. This invention provides two solutions, as follows: Option 1 (High-Performance FPGA Implementation): A Xilinx Zynq-7000 series SoC FPGA is used. The trained 1D-CNN model is compiled into hardware logic circuitry using a High-Level Synthesis (HLS) tool (such as Vitis HLS). Convolutional computations are implemented using parallelized multiply-accumulate (MAC) units, and weights and biases are programmed into the FPGA's Block RAM. This implementation can stably control inference latency to within 1 millisecond.

[0045] Option 2 (High-Performance MCU Implementation): Utilize an NVIDIA Jetson Nano or similar embedded module with a GPU core. Convert the trained model (TensorFlow or PyTorch format) to TensorRT or TensorFlow Lite format and run it on the GPU using its optimized inference engine. While latency is typically 5-10 milliseconds, the development cycle is shorter. To ensure deterministic response, the operating system for the edge nodes should ideally be a Linux distribution pre-patched with the Preempt-RT patch. The rationale for choosing an FPGA or a GPU-enabled MCU is that they offer powerful parallel computing capabilities, making them particularly suitable for running computationally intensive deep learning models. This meets the stringent real-time requirements of industrial environments and avoids latency jitter issues that may occur when running on a general-purpose CPU due to system load fluctuations.

[0046] Specifically, in step S4, the optimization objectives of the global collaborative control engine include at least one of system energy efficiency, device load balancing, and task completion efficiency.

[0047] It should be further explained that the optimization objective of the global collaborative control engine needs to be quantified into a computable reward function to guide the deep reinforcement learning agent. Let the system state at time t be s_t. After the agent executes action a_t, the environment transitions to s_{t+1} and generates a reward r_t. The reward function r_t is a weighted sum of multiple sub-rewards, and its general form is as follows: in: Energy efficiency bonus. Its calculation method is as follows: .here, It is the change in the total power consumption of the system after action a_t is executed (unit: watts). It is a scaling factor (e.g., 0.001) that maps the change in power consumption to an appropriate reward scale. A negative sign indicates that a decrease in power consumption will result in a positive reward; Load balancing rewards. Their calculation method can be based on the CPU utilization of all N computing nodes in the system. The negative of the standard deviation: The smaller the standard deviation, the more balanced the load, and the higher the reward. Task efficiency reward. For example, it can be defined as the number of tasks completed within a time window Δt. ; , , : This is a weighting coefficient, representing the relative importance of each optimization objective. The optimal initial value can be set to... , , Administrators can adjust this according to actual operational strategies.

[0048] Understandably, the reasons for setting weighting coefficients are as follows: in laboratory scenarios, energy efficiency is often a key factor in long-term operating costs, hence its high weighting; load balancing is crucial for equipment lifespan and system stability; and task efficiency directly reflects research output capabilities. This weighted summation method allows system administrators to flexibly adjust the optimization direction.

[0049] Specifically, in step S4, the process of dynamically generating optimized control strategies is based on the continuous interaction between the deep reinforcement learning agent and the system environment, and is obtained through trial and error learning.

[0050] It should be further explained that the core algorithm for dynamically generating policies is the Proximal Policy Optimization (PPO) algorithm, a widely used deep reinforcement learning algorithm known for its training stability. The PPO algorithm involves the following key formulas and steps: 1. Advantage Estimation: Generalized advantage estimation (GAE) is used to more accurately assess the advantage of an action.

[0051] in, It is a timing difference error. It is a discount factor, usually set to 0.99, which indicates the importance attached to future rewards. This is the GAE parameter, usually set to 0.95, used to balance bias and variance.

[0052] 2. Clipped Objective Function of PPO: PPO ensures stability by limiting the magnitude of each policy update. Its objective function is as follows: in, : Parameters of the policy network (Actor network).

[0053] The probability ratio of the new strategy to the old strategy.

[0054] : is the estimated advantage function in formula (1).

[0055] `:` is a hyperparameter representing the clipping range, typically set to 0.2. The rationale for this value is that it allows for effective policy updates while preventing policy collapse caused by excessively large single update steps.

[0056] 3. Value Function Loss: Simultaneously, a Critic network needs to be trained to estimate the state value function. .

[0057] in, These are the parameters of the Critic network. It is a value goal.

[0058] Furthermore, the training process is as follows: The agent interacts in a simulated environment or a secure replica of a real environment, collecting a large number of experience tuples (s_t, a_t, r_t, s_{t+1}). During each training iteration, a mini-batch (size 64) of tuples is randomly sampled from the experience replay buffer, and the Adam optimizer (learning rate 3e-4) is used to minimize the combined loss function. ,in , It is a coefficient. It is the entropy term of the strategy, used to encourage exploration.

[0059] Specifically, the optimization control strategy includes adjusting the operating power of specific devices or migrating computing tasks between devices in different regions.

[0060] It should be further explained that the optimized control strategy is a concretization of the agent's action space A in the PPO algorithm. The action space is designed as a hybrid discrete and continuous action space, specifically as follows: Action 1: Adjust the equipment operating power. This is a continuous action, defined as the proportional coefficient for power adjustment. For example, the agent outputs the action a_t1:(Device_ID_123,k_power=0.8), which means limiting the operating power of the server with device ID 123 to 80% of its maximum power.

[0061] Action 2: Migrate the computation task. This is a discrete action, defined as (Task_ID, Target_Zone_ID). For example, action a_t2:(Task_X, Zone_B) means migrating task X from the current zone to zone B for execution. Task migration involves network data transfer, and its latency d must be considered in the reward function. This can be done through task efficiency rewards. Introduce a penalty term that is negatively correlated with delay, for example ,in It is a small penalty coefficient (e.g., 0.01). This is the total migration latency. In this way, the AI ​​will automatically find a balance between improving efficiency and increasing migration overhead.

[0062] Understandably, the beneficial effect of these specific strategies is that they transform abstract optimization goals into precisely executable control commands, enabling the system's intelligent decision-making to directly generate practical economic benefits such as reduced energy consumption and improved resource utilization.

[0063] Specifically, the method further includes step S6: the cloud platform performs equipment anomaly detection and predictive maintenance alarms based on the standardized equipment data and through a time-series prediction model.

[0064] It should be further explained that step S6 introduces an independent and powerful predictive maintenance layer. A temporal anomaly detection model, such as an LSTM-based autoencoder, is preferred, with the following specific design: 1. Model structure: The encoder is a two-layer LSTM with 64 hidden units in each layer; the decoder is also a two-layer LSTM, and its goal is to reconstruct the input sequence based on the context vector generated by the encoder.

[0065] 2. Training: Unsupervised training was performed using at least 3 months of historical multivariate time-series data (e.g., current, vibration, temperature, with a sampling interval of 1 minute) under normal equipment conditions. The optimization objective was to minimize the reconstruction error (MSE loss), using the Adam optimizer with a learning rate of 1e-3, and training for 100 epochs.

[0066] 3. Online Detection: A standardized device data window of length T=60 (i.e., 1 hour) is input into the trained autoencoder in real-time, and its reconstruction error, `current_error`, is calculated. This error is typically expressed as mean squared error. A dynamic threshold `threshold` is set, which is the 99.5th percentile of the reconstruction errors of all data windows on the training set. The reason for choosing the 99.5th percentile is that it achieves an engineering-acceptable balance between the false positive rate (classifying normal points as anomalous) and the false negative rate (failing to detect anomalies), effectively filtering out most normal fluctuations while capturing significant anomalous patterns. If the reconstruction error of N=3 consecutive data windows exceeds this threshold, a high-level alarm is triggered. At this time, the system calculates an "anomaly confidence score" based on the error value of the last window to quantify the severity of the anomaly. The calculation formula is as follows: in, confidence: Anomaly confidence is a value between 0 and 100, representing the degree of confidence that the system judges the current state to be abnormal. The higher the value, the more significant and certain the anomaly is. current_error: The current reconstruction error is the actual reconstruction error value calculated by the predictive maintenance model based on the input data of the latest time window; threshold: The dynamic detection threshold, which is the 99.5th percentile value defined above and obtained based on historical normal data statistics, is the boundary benchmark for judging normal and abnormal. `min(100, ...)`: This function limits the upper limit of the calculated result to 100. The rationale is that when `current_error` is much larger than `threshold` (e.g., several times the threshold), multiplying this ratio by 100 will result in a value far greater than 100. By "clamping" it at 100, the confidence level can be intuitively represented as a percentage from 0-100%, making it easier for administrators to understand and set trigger conditions for subsequent handling strategies (e.g., triggering an emergency plan when the confidence level > 90%). This approach makes the confidence level metric highly interpretable and practical.

[0067] For example, when current_error is 1.2 times the threshold, the confidence level is min(100, 1.2). The ratio (100) = 120, which, after clamping, is 100, indicating that the system judges it as an anomaly with the highest confidence level. When the ratio is 0.85, the confidence level is 85.

[0068] The beneficial effect of this step is to achieve predictive maintenance, change the operation and maintenance mode from post-event remediation to pre-event early warning, and provide a basis for different levels of alarm response through quantitative confidence, which greatly reduces unplanned downtime and ensures the continuity of laboratory scientific research activities.

[0069] Specifically, when the timing prediction model detects a device anomaly, it triggers the global collaborative control engine to generate a corresponding protective control strategy.

[0070] It should be further explained that this feature enables a deep closed-loop integration of anomaly perception and intelligent control. When the time-series prediction model generates an alarm with a confidence level greater than 90% (e.g., "92% confidence level for motor bearing wear in device A"), it sends a structured event, Alert_Event, to the global collaborative control engine as part of the state s_t. The reward function of the PPO agent immediately introduces a high-priority penalty term. Where C is a large constant (e.g., -10), and I is an indicator function that has a value of 1 when there is an active alarm. This strong negative reward drives the agent to learn quickly and take mitigation measures. Simultaneously, the action space is temporarily augmented with specific protective actions, such as: Reduce_Load(Device_A,0.5): Reduces the load on faulty device A to 50%.

[0071] Switch_To_Backup(Device_A): Switch to the backup device for device A.

[0072] Isolate_Device(Device_A): Isolate device A under safe conditions.

[0073] Understandably, intelligent agents learn by interacting with their environment. Driven by this mechanism, the optimal combination of protective actions is selected to eliminate or mitigate the impact of failures. The beneficial effect of this linkage is that it endows the system with a high degree of autonomy and resilience, enabling it to proactively respond to equipment failures and maximize the availability and safety of the entire laboratory system.

[0074] Please see Figure 2 The present invention provides another embodiment, which provides a cross-regional laboratory management system based on protocol adaptive parsing. The cross-regional laboratory management system based on protocol adaptive parsing includes: Edge nodes 100 are deployed in various laboratory areas. Each edge node is equipped with a protocol adaptive parsing module, which is used to automatically identify and parse the communication data stream of the access device and generate standardized device data. The cloud platform 200 is communicatively connected to all the edge nodes 100. The cloud platform is equipped with a global collaborative control engine, which is used to dynamically generate cross-regional optimized control strategies based on the received standardized device data and distribute them to the corresponding edge nodes. The protocol adaptive parsing module is based on a deep learning model, and the global collaborative control engine is based on a deep reinforcement learning model.

[0075] It should be further explained that this system is the physical implementation of the aforementioned method. The edge node hardware uses Advantech ARK-3500 series industrial-grade fanless PCs, equipped with an Intel Core i7 processor and an M.2 interface Intel Arria 10GX FPGA acceleration card for running the protocol parsing model. Lightweight container runtimes (such as Docker) are deployed on the nodes, running microservices such as data acquisition, protocol parsing, and local control. The cloud platform is built on a Kubernetes cluster, using a MySQL database to store historical data and Redis as a real-time cache. The global collaborative control engine, predictive maintenance service, and web management portal are all deployed as containerized microservices. Network isolation between the edge and the cloud is ensured through Calico network policies, and communication security is ensured using mutual TLS authentication (mTLS). Internal communication between all services uses the gRPC protocol to guarantee high performance and interface consistency.

[0076] Understandably, the beneficial effect of this system is that it provides a high-performance, highly reliable, and horizontally scalable integrated hardware and software solution that closely combines advanced algorithm theory with industrial-grade engineering practices, ensuring that the intelligent management method can operate stably 24 / 7 in complex real-world laboratory environments.

[0077] In a preferred embodiment, this application also provides an electronic device, the electronic device comprising: The computer device includes a memory and a processor, wherein the memory stores computer-readable instructions that, when executed by the processor, implement the protocol-adaptive parsing-based cross-regional laboratory management method described above. The computer device can be broadly categorized as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the computer device can be used to provide the necessary computing, processing, and / or control capabilities. The memory of the computer device may include non-volatile storage media and internal memory. The non-volatile storage media may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface and communication interface of the computer device can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the method of the present invention.

[0078] This invention can be implemented as a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the steps of the methods of embodiments of the invention to be performed. In one embodiment, the computer program is distributed across multiple network-coupled computer devices or processors, such that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, may be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations may be executed by one or more computer devices or processors, and one or more other method steps / operations may be executed by one or more other computer devices or processors. One or more computer devices or processors may execute a single method step / operation, or execute two or more method steps / operations.

[0079] Those skilled in the art will understand that the method steps of this invention can be performed by a computer program instructing related hardware, such as a computer device or processor, to perform the steps of this invention when executed. Depending on the context, any references herein to memory, storage, databases, or other media may include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0080] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0081] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A cross-regional laboratory management method based on protocol adaptive parsing, characterized in that, Includes the following steps: S1. Edge nodes deployed in various laboratory areas monitor the communication data streams of newly connected devices in real time. S2. The communication data stream is automatically identified and parsed by the protocol adaptive parsing module based on a deep learning model to generate standardized device data. S3. Upload the standardized equipment data to the cloud platform; S4. Through the cloud platform's global collaborative control engine based on deep reinforcement learning, standardized device data received from multiple edge nodes is analyzed to dynamically generate cross-regional optimized control strategies. S5. The optimized control strategy is sent to the corresponding edge nodes for execution.

2. The method as described in claim 1, characterized in that, In step S2, the deep learning model is a one-dimensional convolutional neural network or a Transformer sequence model.

3. The method as described in claim 2, characterized in that, In step S2, the specific process of automatic identification and parsing includes: extracting the timing features and frame structure features of the communication data stream, and outputting the parsed device parameters and status information.

4. The method as described in claim 3, characterized in that, The protocol adaptive parsing module runs on the field-programmable gate array or microprocessor of the edge node to ensure the real-time performance of protocol parsing.

5. The method as described in claim 1, characterized in that, In step S4, the optimization objectives of the global collaborative control engine include at least one of system energy efficiency, device load balancing, and task completion efficiency.

6. The method as described in claim 5, characterized in that, In step S4, the process of dynamically generating optimized control strategies is based on the continuous interaction between the deep reinforcement learning agent and the system environment, and is obtained through trial and error learning.

7. The method as described in claim 6, characterized in that, The optimization control strategy includes adjusting the operating power of specific devices or migrating computing tasks between devices in different regions.

8. The method as described in claim 1, characterized in that, The method further includes step S6: the cloud platform performs equipment anomaly detection and predictive maintenance alarms based on the standardized equipment data and through a time-series prediction model.

9. The method as described in claim 8, characterized in that, When the timing prediction model detects a device malfunction, it triggers the global collaborative control engine to generate a corresponding protective control strategy.

10. A cross-regional laboratory management system based on protocol adaptive parsing, characterized in that, The system for implementing the method as described in any one of claims 1 to 9 includes: Edge nodes deployed in various laboratory areas are equipped with a protocol adaptive parsing module, which is used to automatically identify and parse the communication data streams of access devices and generate standardized device data; The cloud platform communicates with all the edge nodes. The cloud platform is equipped with a global collaborative control engine, which is used to dynamically generate cross-regional optimized control strategies based on the received standardized device data and distribute them to the corresponding edge nodes. The protocol adaptive parsing module is based on a deep learning model, and the global collaborative control engine is based on a deep reinforcement learning model.

Images