Ocean buoy electrical appliance collection and beidou communication fault-tolerant transmission method, medium and system

Abstract

The application provides a kind of ocean buoy electric appliance collection and fault-tolerant transmission method, medium and system of communication, belong to communication technical field, the application is by establishing sensor task table and using dynamic potential field sensor time-sharing polling dynamic flow scheduling model driven time-sharing polling collection, executes three levels quality determination to collection data and packs transmission, uses energy functional physics constraint jump cycle prediction model output battery remaining capacity and health state to execute adaptive sampling period adjustment and branch load power supply control, by the incremental merging of storage control unit and sector alignment write-in mechanism storage data and maintain pending queue, by beidou communication machine uploads shore end and after communication recovers according to priority, make up simultaneously, to fault sensor executes power-off retry and isolation processing, solve the data frame collision when multiple sensors concurrent access bus, and the state evaluation result of battery deviates from the actual physical law under non-standard working condition artificial intelligence technical problem.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, and specifically relates to a method, medium, and system for electronic data acquisition and BeiDou communication fault-tolerant transmission of marine buoys. Background Technology

[0002] In the field of marine environmental monitoring, buoy systems typically employ a main control unit that uses an RS485 bus to poll multiple sensors in a time-division multiplexing manner. The remaining battery power is determined based on a simple threshold value for the battery terminal voltage to control sensor power supply. This approach is widely used in simultaneous observation scenarios for multiple parameters, such as hydrology, meteorology, and water quality. However, when multiple sensors simultaneously reach their acquisition time, or when the battery is in a non-steady-state charging and discharging condition, the fixed-priority polling strategy and simple voltage threshold judgment are difficult to adapt to dynamically changing bus occupancy and battery aging. In other words, existing technologies suffer from technical problems such as data frame collisions easily occurring when multiple sensors concurrently access the bus, and battery status assessment results deviating from actual physical laws under non-standard operating conditions. Summary of the Invention

[0003] In view of this, the present invention provides a method, medium and system for fault-tolerant transmission of electrical data acquisition and BeiDou communication for marine buoys, which can solve the technical problem in the prior art that it is difficult to balance the orderliness of bus access and the accuracy of energy management in the process of concurrent data acquisition by multiple sensors and energy status assessment of batteries for marine buoys.

[0004] The present invention is implemented as follows: The first aspect of the present invention provides a method for fault-tolerant transmission of marine buoy electrical data acquisition and BeiDou communication, comprising the following steps:

[0005] A sensor task table is established for each sensor. Based on the dynamic flow scheduling model of time-sharing polling for dynamic potential energy field sensors, each acquisition task is scheduled, driving the main control unit to execute the time-sharing polling acquisition of each sensor in sequence according to time slices.

[0006] The main control unit performs three levels of quality judgment on the collected data: communication level, physical quantity level, and trend level. It adds data quality flag bits and fault codes to each frame of data and packages the data into a unified data packet.

[0007] The energy functional physics constraint jump cycle prediction model is based on real-time collected battery terminal voltage, charging and discharging current and temperature, and outputs the remaining battery capacity and battery health status. The main control unit performs adaptive sampling period adjustment and branch load power supply control according to the remaining battery capacity and battery health status.

[0008] The main control unit sends data packets to the storage control unit, which writes the data to the solid-state drive through incremental merging of the memory log buffer and sector-aligned writing mechanism, and maintains the queue to be sent and the sending status.

[0009] The main control unit uploads the data packet to the shore end through the Beidou communication device. If no confirmation is received from the shore end, the data packet is kept in the state of waiting to be sent. After communication is restored, the cached resending is performed according to the principles of alarm priority, recent priority, and sequential resending.

[0010] When the number of consecutive unresponsive times of the sensor reaches the fault judgment threshold, the main control unit executes the process of power-off, delayed power-on, and retry data acquisition. If there is still no response after the retry, the sensor is placed in a fault isolation state and a fault code is uploaded to the shore. After the shore sends a command, the main control unit verifies the command and performs query, parameter modification, reset, or historical data retransmission operations.

[0011] The main control unit is an STM32 microcontroller.

[0012] The sensor task table includes sensor number, interface type, power supply channel number, sampling period, communication command content, response timeout duration, maximum number of retries, data parsing function pointer, normal range upper and lower limits, and rate of change threshold.

[0013] The dynamic flow scheduling model of the dynamic potential energy field sensor in time-sharing polling abstracts the urgency of acquisition as the mass of a point mass, and abstracts interface conflicts and power consumption limits as a virtual resistance potential energy field.

[0014] Among them, conflicting tasks on the shared bus are pushed into the waiting queue due to virtual charge repulsion, and the potential energy field damping coefficient increases when the total power consumption of the system approaches the safety threshold.

[0015] Specifically, the communication-level quality assessment verifies the consistency between the frame header identifier, the frame length field and the actual data length, and the cyclic redundancy check value.

[0016] The physical quantity quality determination and the trend level quality determination are respectively range comparison and adjacent sampling difference comparison.

[0017] The energy functional physical constraint jump cycle prediction model is based on a multi-layer recurrent neural network structure, which includes a feature embedding fully connected layer and a two-layer recurrent neural network encoder.

[0018] The recurrent neural network encoder is equipped with a dynamic gating jump mechanism, which activates the step jump function or fine-tunes the loop branch based on the rate of change of the terminal voltage.

[0019] The energy functional physics constraint jump cycle prediction model parallel capacity decay path selection submodule uses a graph theory shortest path algorithm to filter decay paths.

[0020] The loss function of the energy functional physical constraint jump cycle prediction model includes a data fitting error term and a physical constraint regularization term, and the weight coefficient of the physical constraint regularization term is dynamically adjusted by the model accuracy control function.

[0021] Specifically, the adaptive sampling period adjustment is based on the battery's remaining capacity, health status, terminal voltage, and minimum operating voltage of the load, which are used to set high voltage threshold, low voltage threshold, and minimum protection voltage threshold to divide the sampling mode.

[0022] Specifically, the incremental merging and sector-aligned writing mechanism of the memory log buffer merges data packets and writes them in batches according to integer multiples of the logical block size reported by the storage device and the file system block size.

[0023] The queue to be sent includes a unique sequence number of the data packet, a collection timestamp, a data type, a data priority, a number of times it has been sent, a last time it was sent, a sending status flag, and a checksum.

[0024] The fault determination threshold is defined as 3 to 5 consecutive communication failures, with a default setting of 3 consecutive communication failures. After any successful communication attempt, the consecutive communication failure count is reset to zero. When the consecutive communication failure count reaches the fault determination threshold, the main control unit cuts off the power supply to the corresponding sensor, delays for 2000 to 5000 milliseconds, and then re-energizes it, with a default delay of 3000 milliseconds. The retry acquisition time after re-energizing is no earlier than the startup stabilization time specified in the sensor's technical manual. If the startup stabilization time exceeds 5000 milliseconds, the time specified in the sensor's technical manual shall prevail.

[0025] The high voltage threshold, low voltage threshold, and minimum protection voltage threshold are calibrated based on the battery's chemical system, number of cells in series, temperature, load current, health status, remaining capacity output by the BMS, and the minimum allowable operating voltage of the critical load, rather than being determined as a fixed percentage of the battery's nominal voltage. Taking a deep-cycle lead-acid battery with a rated voltage of 12V as an example, under calibration conditions of an ambient temperature of 25±1℃ and the battery being stopped from charging and discharging and left to stand for at least 6 hours, the high voltage threshold is set to 12.50V, the low voltage threshold is set to 12.10V, and the minimum protection voltage threshold is set to 11.66V; when the battery terminal voltage is lower than 11.66V for three consecutive sampling cycles, the main control unit enters the system protection mode.

[0026] A second aspect of the present invention provides a computer-readable storage medium storing program instructions, which, when executed in a computer, are used to perform the above-described method for the electrical data acquisition and BeiDou communication fault-tolerant transmission of marine buoys.

[0027] A third aspect of the present invention provides an electrical data acquisition and BeiDou communication fault-tolerant transmission system for marine buoys, comprising the aforementioned computer-readable storage medium, wherein the system is a computer, the computer-readable storage medium is disposed within the system, and the system is provided with a microprocessor for executing program instructions stored in the computer-readable storage medium.

[0028] This invention addresses the technical problems of data frame collisions during concurrent bus access by multiple sensors and battery condition assessment results deviating from actual physical laws under non-standard operating conditions by combining a dynamic flow scheduling model based on time-sharing polling of a kinetic potential energy field sensor with a jump-loop prediction model based on energy functional physics constraints. This invention abstracts bus occupancy conflicts into a virtual resistance potential energy field, ensuring the scheduling decisions of the acquisition tasks have continuous evolutionary physical consistency, thus avoiding bus contention under fixed priority strategies at the mechanism level. Simultaneously, it incorporates electrochemical polarization and thermodynamic degradation laws into the neural network loss function constraints, ensuring that battery condition prediction results always conform to physical evolution laws, avoiding distorted predictions by purely data-driven models under abnormal operating conditions. In summary, this invention solves the technical problems mentioned in the background art, namely, data frame collisions during concurrent bus access by multiple sensors and battery condition assessment results deviating from actual physical laws under non-standard operating conditions. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method of the present invention.

[0030] Figure 2 A distribution of bus wait times for sensor tasks with different weights.

[0031] Figure 3 This is a comparison diagram of the battery terminal voltage sampling mode and the branch power supply status.

[0032] Figure 4 This is a schematic diagram of the hardware system involved in this embodiment. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0034] like Figure 1 The diagram shown is a flowchart of a method for fault-tolerant transmission of marine buoy electrical data acquisition and BeiDou communication provided by the first aspect of this invention. This method includes the following steps:

[0035] S01. Establish a sensor task table for each sensor, schedule each acquisition task based on the dynamic flow scheduling model of the dynamic potential energy field sensor time-sharing polling, and drive the main control unit to execute the time-sharing polling acquisition of each sensor in sequence according to the time slice;

[0036] S02. The main control unit performs three-level quality judgment on the collected data: communication level, physical quantity level, and trend level. It adds data quality flag bits and fault codes to each frame of data and packages the data into a unified data packet.

[0037] S03, the energy functional physical constraint jump cycle prediction model is based on the real-time collected battery terminal voltage, charging and discharging current and temperature, and outputs the remaining battery capacity and battery health status. The main control unit performs adaptive sampling period adjustment and branch load power supply control according to the remaining battery capacity and battery health status.

[0038] S04. The main control unit sends the data packet to the Allwinner V40 storage control unit. The Allwinner V40 storage control unit writes the data to the solid-state drive through the incremental merging and sector-aligned writing mechanism of the memory log buffer, and maintains the queue to be sent and the sending status.

[0039] S05. The main control unit uploads the data packet to the shore end through the Beidou communication device. If no confirmation is received from the shore end, the data packet is kept in the state of waiting to be sent. After the communication is restored, the cached resending is performed according to the principle of alarm priority, recent priority, and sequential resending.

[0040] S06. When the number of consecutive unresponsive times of the sensor reaches the fault judgment threshold, the main control unit executes the circuit power-off, delayed power-on, and retry acquisition process. If there is still no response after retry, the sensor is set to fault isolation state and a fault code is uploaded to the shore. After the shore sends the command, the main control unit verifies the command and performs query, parameter modification, reset, or historical data retransmission operations.

[0041] The main control unit is an STM32 microcontroller.

[0042] The sensor task table is a pre-established structured array within the main control unit. Each entry contains at least the sensor number, interface type, power supply channel number, sampling period, communication command content, response timeout duration, maximum number of retries, data parsing function pointer, normal range upper and lower limits, and rate of change threshold. It is used to uniformly describe the acquisition parameters and anomaly judgment conditions of all sensors in the system, avoiding interface conflicts and parameter inconsistencies caused by scattered acquisition logic.

[0043] The dynamic potential energy field sensor time-sharing polling dynamic flow scheduling model is a scheduling algorithm that maintains a dynamic potential energy matrix within the main control unit. It abstracts the urgency of data acquisition from each sensor into the mass of a virtual mass point, and abstracts interface conflicts, bus occupancy, and power consumption limits into a virtual resistance potential energy field. When the timer triggers the sampling period, it solves the simplified mass point dynamic equation in real time and calculates the resultant force on each acquisition task in the potential energy field. High-priority tasks and tasks nearing their acquisition deadlines gain priority access to the bus due to increased virtual gravity. Conflicting tasks sharing the same RS485 bus are pushed into the waiting queue due to virtual charge repulsion. When the total system power consumption approaches the safety threshold, the potential energy field damping coefficient automatically increases, thereby lengthening the sampling interval of non-critical tasks. This physically mitigates instantaneous power consumption and fundamentally avoids access conflicts on the RS485 half-duplex bus. The safety threshold for the potential energy field damping coefficient was determined through multiple rounds of comparative experiments and data analysis using an equivalent test platform with the same sensor configuration as the target buoy, built in the laboratory. The peak instantaneous power consumption of each sensor when simultaneously powered on was recorded, and combined with the maximum continuous discharge current limit of the battery. The power consumption safety threshold was set to the minimum allowable continuous power of the battery or BMS, DC / DC converter, cables, and switching devices, with an engineering margin. 75%–85% was used only as an initial value for joint debugging; the final value was determined by temperature rise and voltage drop tests of the target hardware. The mass assignment rule was based on the urgency of sensor acquisition, assigning integer priority weights from high to low. It is recommended that the weight range for alarm sensors be 8–10, for power status sensors 6–8, for hydrological sensors 4–6, and for meteorological sensors 2–4. Specific values ​​were determined after analyzing the acquisition delay sensitivity of the buoy's actual operating data. The technical benefits of the dynamic flow scheduling model for the dynamic potential energy field sensor time-sharing polling are as follows: This algorithm maps the abstract multi-task scheduling problem into a dynamic evolution process in a continuous potential energy field, making the scheduling decision physically continuous and self-consistent. The mutual exclusion relationship between tasks is naturally encoded through the repulsive force field, and combined with the bus occupancy flag or interface token, it ensures that only one task is sent on the same RS485 bus at the same time. This eliminates the risk of data frame collisions caused by multiple sensors accessing the RS485 half-duplex bus at the mechanism level. At the same time, the dynamic adjustment of the damping coefficient realizes the coordinated optimization of sampling frequency and system power consumption, so that the buoy can still maintain a stable energy balance in the scenario of concentrated deployment of high-power sensors.

[0044] The communication-level quality assessment refers to verifying the consistency between the frame header identification, the frame length field and the actual data length, and the CRC cyclic redundancy check value of the acquired frames. The physical quantity-level quality assessment refers to comparing the parsed physical quantity values ​​with the preset upper and lower limits of the range in the sensor task table. The trend-level quality assessment refers to comparing the difference between two adjacent sampled values ​​with the preset rate of change threshold in the sensor task table. The rate of change threshold is determined by statistically analyzing historical observation data of the target sea area, extracting the maximum rate of change of each physical quantity under normal and abnormal sea conditions, using the high quantile value or robust statistics of the historical effective data rate of change as the initial threshold, and calibrating it in conjunction with sensor accuracy and on-site false alarm rate. The data quality flag is a status field attached to the unified data packet, with values ​​representing four states: normal communication, abnormal communication verification, physical quantity out of range, and abnormal trend change, used to transmit data reliability information to the shore. The fault code is an encoded field embedded in the unified data packet, used to identify specific fault types such as sensor communication timeout, interface verification failure, power supply undervoltage, and storage write abnormality, facilitating remote fault location by the shore.

[0045] The energy functional physics constraint jump loop prediction model is a deep learning-based prediction model for battery remaining capacity and health status. Its specific structure is as follows: The model is based on a multi-layer recurrent neural network. The input layer receives time-series data of battery terminal voltage, charging / discharging current, and temperature organized at fixed sampling intervals. The recommended time-series window length is 20–60 time steps, with each time step having a feature dimension of 3. A fully connected feature embedding layer follows the input layer, mapping the original three-dimensional time-series features to a high-dimensional latent space, with a recommended latent space dimension of 32–64. The latent space features are input to the core two-layer recurrent neural network encoder, with each layer containing a latent state dimension. The recommended value is 32-48. Within each loop layer, a dynamic gating jump mechanism based on the rate of change of terminal voltage is designed. When the absolute value of the rate of change of terminal voltage in adjacent time steps is lower than the stable discharge threshold, the step jump function is activated, skipping redundant matrix multiplication operations of multiple hidden states in the middle, and directly passing the previous valid hidden state to the current step through a weighted residual connection, achieving fast iteration across step lengths. When the absolute value of the rate of change of terminal voltage exceeds the stable discharge threshold, the step state is immediately interrupted and jumps to the fine-tuning loop branch. The fine-tuning loop branch is an independent single-layer recurrent neural network, with a recommended hidden state dimension of 16-24, used for step-by-step processing of high-current charging / discharging or instantaneous undervoltage segments. The model employs high-precision iteration; the dual-branch outputs are weighted and summed at the fusion layer using learnable concatenated weight vectors. After fusion, a single-layer fully connected output head outputs the predicted values ​​of the remaining battery capacity and battery health status, respectively. Between the hidden and output layers, a capacity degradation path selection submodule based on a graph theory shortest path algorithm is connected in parallel. Different charge / discharge historical paths are abstracted as directed graph nodes. The health status degradation rate under each path is used as the edge weight. The Dijkstra algorithm is used to filter the degradation path that best matches the current battery historical state in real time during the inference phase. The degradation slope of the selected path is injected as an additional bias into the output layer, thereby ensuring that the prediction results are accurate and accurate. To maintain physical consistency with the battery aging stage, the model's loss function consists of two parts: a data fitting error term and a physical constraint regularization term. The physical constraint regularization term includes constraints from the electrochemical polarization equation and thermodynamic degradation physical laws. The electrochemical polarization equation constraint requires that the deviation between the predicted terminal voltage and the estimated terminal voltage calculated based on the equivalent circuit model be penalized. The thermodynamic degradation physical law constraint requires that the battery health state sequence maintain a monotonically non-increasing physical law in the time dimension. If the predicted health state sequence shows a local rebound, an additional penalty weight is applied. The weight matrix between neurons in each layer of the model adopts a sparse connection initialization strategy, and the sparsity is recommended to be 0.2 to 0.4. To reduce the amount of multiplication and accumulation operations during the inference stage of the embedded chip; the activation function between layers adopts a piecewise linear approximation to replace the standard hyperbolic tangent function, transforming floating-point multiplication operations into shift addition operations, thereby reducing the computational resource consumption of the embedded chip; the transfer of implicit states of data within the loop layer along the time axis adopts fixed-point quantization representation, with a recommended quantization bit width of 8 to 12 bits to balance inference accuracy and embedded chip memory consumption; the overall number of model parameters is recommended to be controlled between 30,000 and 80,000 to meet the memory constraints of deployment on the main control unit or Allwinner V40 memory control unit. The steps for establishing the training dataset for the energy functional physical constraint jump cycle prediction model specifically include: using a battery of the same model as the actual configuration of the buoy in a laboratory environment, cyclically charging and discharging under various operating conditions such as constant current discharge, pulse discharge, and discharge simulating the actual load curve of the buoy using a battery comprehensive tester, and simultaneously collecting time-series data of terminal voltage, charging and discharging current, and temperature, and labeling the true value of the remaining battery capacity for each charge and discharge cycle using the Coulomb integral method; as the number of cycles increases, recording the capacity decay curve, using 80% of the initial rated capacity as the boundary for determining the termination of the healthy state, and establishing a sample set covering different aging stages throughout the entire life cycle; in addition, extracting battery electrical parameter records from historical data collected by buoys already in operation in the target sea area as supplementary samples to expand the coverage of real marine environmental operating conditions in the dataset; finally, dividing all samples into training set, validation set, and test set in a ratio of 8:1:1. The specific steps for training the energy functional physical constraint jump loop prediction model include: using mean squared error as the data fitting error term, and using electrochemical polarization equation bias penalty and health state monotonicity penalty as physical constraint regularization terms, the three terms are weighted and summed to form the total loss function. The initial weight coefficient of the physical constraint regularization term is recommended to be set to 0.1 to 0.3, and dynamically adjusted during training through the model accuracy control function; the Adam optimizer is used for parameter updates, and the initial learning rate is recommended to be set to a certain value. ~ The recommended total number of training rounds is 100 to 300. After each round, the mean squared error should be evaluated on the validation set. If the validation set error does not decrease for 10 to 20 consecutive rounds, the early stopping mechanism should be triggered. After training is completed, the model weight parameters should be quantized to the target bit width, and the accuracy loss after quantization should be verified on the test set to see if it is within an acceptable range. The technical benefits of the energy functional physical constraint jump-loop prediction model are as follows: This model directly incorporates the electrochemical polarization mechanism and thermodynamic degradation law into the loss function constraint, so that the recurrent neural network is subject to the forced constraint of physical laws in addition to its time-series modeling capabilities. Therefore, even when the battery is under non-standard discharge conditions or the sensor has slight drift, the prediction results can still maintain the trend of change that conforms to physical laws, avoiding the generation of physically impossible prediction values ​​on out-of-distribution samples by the purely data-driven model. The dynamic gating jump mechanism enables the model to significantly reduce redundant matrix operations and reduce the inference latency of embedded chips during the stable discharge period, while automatically switching to a fine iteration mode during the violent charge and discharge period to ensure the prediction accuracy at critical moments. This achieves the dynamic allocation of computing resources and the synergistic optimization of prediction accuracy, providing physically reliable energy state support for adaptive sampling and load control.

[0046] The model accuracy adjustment function is used to dynamically adjust the weight coefficients of the physical constraint regularization term during the training of the energy functional physical constraint jump loop prediction model. The function calculates a comprehensive adjustment index based on three indicators: the mean square error of the current validation set data fitting, the physical constraint violation rate, and the mean absolute error of the remaining battery capacity prediction. When the comprehensive adjustment index belongs to... When the physical constraints of the model are too strong, it indicates that the fitting ability is limited. Therefore, the weight coefficients of the physical constraint regularization are reduced by a fixed step size. When the comprehensive control index belongs to... When the model is in a balance between accuracy and physical consistency, the current weight coefficients are maintained; when the comprehensive control index belongs to When the physical constraints of the model are insufficient, the prediction results will show physical violations. The weight coefficient of the physical constraint regularization will be increased by a fixed step size. The segment boundaries of the comprehensive control index are not preset to be fixed at 0.3 and 0.7, but are determined by the Pareto front inflection point of the prediction accuracy and the physical constraint satisfaction rate on the validation set.

[0047] The adaptive sampling period adjustment refers to the main control unit setting three graded control boundaries—high voltage threshold, low voltage threshold, and minimum voltage threshold—based on the battery's remaining capacity and health status output by the energy functional physical constraint jump cycle prediction model, combined with the real-time battery terminal voltage. When the battery terminal voltage is not lower than the high voltage threshold, it enters normal sampling mode; when the battery terminal voltage is between the low and high voltage thresholds, it enters energy-saving sampling mode; when the battery terminal voltage is between the minimum voltage threshold and the low voltage threshold, it enters minimum operating mode; and when the battery terminal voltage is lower than the minimum voltage threshold, it enters system protection mode. Each voltage threshold is calibrated based on the battery chemical system, number of cells in series, remaining capacity output by the BMS, minimum single-cell voltage, temperature, load current, and minimum operating voltage of critical loads, and is not uniformly converted to a fixed percentage of the nominal voltage. For a rated 12V sensor, 11V can only continue to operate when the device's allowable input lower limit is not higher than 11V and the power supply is stable; otherwise, it enters energy-saving or protection mode. The aforementioned branch load power supply control refers to the main control unit controlling the independent power supply circuits of each sensor and communication device through relays or electronic switches, and sequentially shutting down or restoring the power supply to non-critical sensors according to load priority under different sampling modes.

[0048] The incremental merging and sector-aligned write mechanism of the memory log buffer refers to the Allwinner V40 storage control unit maintaining a log buffer in memory. Data packets from the control unit are first written to the memory buffer. When the amount of data in the buffer reaches an integer multiple of the physical sector size of the solid-state drive, an alignment write operation is performed, writing the buffer data to the solid-state drive in batches by sector. This merges multiple small data writes into a single large block write with sector alignment, reducing the physical erase and write frequency of the solid-state drive and reducing uneven block wear. The buffer trigger write amount is aligned to the common multiple of the logical block size reported by the storage device and the file system block size. When the target device reports an alignment unit of 4 KiB, it can be determined through joint debugging at integer multiples such as 16 KiB or 32 KiB, and 4096 bytes or 4 to 8 times it is not used as a fixed parameter for all SSDs.

[0049] The queue to be sent is a structured index list maintained by the Allwinner V40 storage control unit on the solid-state drive. Each record contains at least a unique sequence number of the data packet, a collection timestamp, data type, data priority, number of transmissions, last transmission time, transmission status flag, and CRC checksum. This queue is used to save the status of data packets not confirmed by the shore during BeiDou communication anomalies and to support retransmission scheduling based on priority and time order after communication is restored. The principle of alarm priority, recent priority, and sequential retransmission means that after communication is restored, the retransmission scheduler first selects data packets with alarm data type, then selects the latest real-time data packets with the most recent timestamp, and finally selects ordinary historical data packets in ascending order of timestamp. These are then uploaded sequentially through the BeiDou communication device and await confirmation from the shore before updating the transmission status flag.

[0050] The fault determination threshold is the number of consecutive communication failures that trigger the sensor shunt power-off and fault isolation process. Its value is calibrated based on the target sensor's normal communication frame error rate, allowable missed detection rate, and on-site interference level, and is configured separately in the sensor task table. The delayed power-on duration is not less than the power-on stabilization time specified in the corresponding sensor technical manual, avoiding the use of a uniform, fixed range to cover different devices.

[0051] The hardware system involved in this method mainly consists of an STM32 microcontroller main control unit as the core scheduling hub, an Allwinner V40 storage control unit responsible for high-reliability storage, a power management and power supply control module, a Beidou communication unit, and a multi-source sensor array. In terms of electrical connections, the STM32 microcontroller's multiple GPIO pins are electrically connected to the RS485 bus transceiver chip and the control terminals (such as the base of relay-driven transistors or the enable terminals of electronic switches) of each branch load in the power management module. Its built-in ADC sampling pin is electrically connected to the battery's terminal voltage, charge / discharge current detection circuit, and temperature sensor. The STM32 and Allwinner V40 are bidirectionally connected via SPI or UART buses. The Allwinner V40's storage control bus is physically connected to the solid-state drive (SSD) board level, while its communication serial port is electrically connected to the baseband data interface of the Beidou communication unit. The Beidou communication unit's RF output terminal is interconnected with an external Beidou antenna via a coaxial cable, thus constructing a closed-loop hardware topology integrating data acquisition, storage, power supply, and communication.

[0052] In terms of device function and role, the STM32 main control unit utilizes an internal sensor task table to run a dynamic potential energy field scheduling algorithm to smooth instantaneous power consumption and eliminate bus conflicts. Simultaneously, it performs a three-level quality assessment on the acquired data and encapsulates fault codes. Furthermore, it runs an energy functional prediction model to evaluate the battery health status, thereby driving GPIO to execute adaptive branch power supply control and sensor fault power-off reset protection. The Allwinner V40 storage control unit opens a memory log buffer, using incremental merging and sector-aligned write mechanisms to batch write data to the SSD to reduce physical erase / write frequency. It also maintains a queue of data to be transmitted on the SSD, scheduling BeiDou retransmission according to alarm and recent priority principles after communication is restored. The power management and power supply control module, driven by the STM32, acts as a physical actuator to cut off or connect designated branch power supplies to achieve energy saving or hardware reset. Finally, the BeiDou communication unit uses the satellite network to achieve all-weather, high-fault-tolerant bidirectional data wireless radio frequency transmission between the buoy and the land shore.

[0053] The specific implementation of step S01 involves first establishing a structured array-based sensor task table in the main control unit's storage area, recording parameters such as the interface type, power supply channel number, sampling period, and parsing function pointer for each sensor to uniformly describe the acquisition conditions. Then, a dynamic potential energy matrix is ​​maintained within the main control unit, mapping the urgency of each sensor acquisition task to a virtual mass, and mapping bus occupancy and interface conflicts to a virtual drag potential energy field. At each timer trigger, the resultant force on each acquisition task in the potential energy field is calculated in real time. Tasks with high urgency or nearing their acquisition deadline gain priority access to the bus due to increased virtual attraction, while conflicting tasks sharing the same bus are pushed into a waiting queue due to repulsion. When the total system power consumption approaches a preset safety threshold, the potential energy field damping coefficient automatically increases, thereby lengthening the sampling interval for non-critical tasks. The purpose of this step is to transform the discrete multi-task scheduling problem into a continuous physical evolution process, making bus access inherently mutually exclusive. The power consumption safety threshold is taken as the minimum value among the allowable continuous power of the battery or BMS, DC / DC converter, cables and switching devices, with engineering margins; 75% to 85% is only used as the initial value for joint debugging.

[0054] The specific implementation of step S02 involves sequentially performing three levels of quality assessment—communication level, physical quantity level, and trend level—on each frame of raw data received by the main control unit. The communication level assessment verifies the consistency between the frame header identifier, the frame length field, and the actual data length, as well as the cyclic redundancy check value, to determine if the data frame is complete during transmission. The physical quantity level assessment compares the parsed values ​​with the preset upper and lower limits of the range in the sensor task table to determine if the values ​​are within a reasonable range. The trend level assessment compares the difference between two adjacent sampled values ​​with a preset rate of change threshold to determine if the numerical change conforms to normal fluctuation patterns. The rate of change threshold is determined using the high quantile value of the historical effective data rate of change or a robust statistic, and is calibrated in conjunction with sensor accuracy and the on-site false alarm rate. After the assessment is completed, a data quality flag bit containing four states—communication normal, communication verification abnormal, physical quantity out of range, and trend change abnormal—is added to each frame of data, and a fault code for identifying the specific fault type is embedded. Finally, the data is packaged together.

[0055] The specific implementation of step S03 involves organizing the real-time collected battery terminal voltage, charging / discharging current, and temperature into a time-series window at fixed sampling intervals, and inputting it into the energy functional physical constraint jump-loop prediction model. This model employs a multi-layer recurrent neural network structure. First, a feature embedding fully connected layer maps the three-dimensional time-series features to the latent space, and then the data is input into a two-layer recurrent neural network encoder. The encoder internally incorporates a dynamic gating jump mechanism. When the terminal voltage change rate is lower than the stable discharge judgment threshold, a step jump function is activated to skip redundant calculations; when the change rate exceeds the threshold, the system switches to a fine-tuning loop branch for point-by-point iteration. The dual-branch outputs are weighted and summed in the fusion layer, and then output as the remaining battery capacity and battery health status via a fully connected output head. The model's parallel capacity degradation path selection submodule uses a graph theory shortest path algorithm to select the degradation path that best matches the current battery history during the inference phase, and injects the path degradation slope as a bias into the output layer. The main control unit, based on the remaining output capacity and health status, and combined with the real-time terminal voltage, divides the voltage into three levels: high voltage threshold, low voltage threshold, and minimum protection voltage threshold. It then enters the corresponding sampling mode and executes branch load power supply control via relays or electronic switches. Specific thresholds are calibrated according to the actual battery system, BMS parameters, and the minimum operating voltage of the load, and are not uniformly determined using a fixed percentage of the nominal voltage. For a rated 12V sensor, 11V is only required for continued operation when the device's allowable input range covers 11V.

[0056] The specific implementation of step S04 involves the main control unit sending the packaged data packets to the storage control unit via the communication bus. The storage control unit allocates a log buffer in memory, temporarily storing the received data packets in the buffer. When the buffer size reaches an integer multiple of the physical sector size of the solid-state drive (SSD), an aligned write operation is performed, writing the data to the SSD in batches, sector by sector. This mechanism merges multiple small data writes into a single batch write. The buffer-triggered write size is aligned to the least common multiple of the logical block size reported by the target storage device and the file system block size, determined through testing. After the write is complete, the storage control unit maintains a queue to be sent on the SSD, recording the unique sequence number, timestamp, data type, priority, number of transmissions, and transmission status flag for each data packet.

[0057] The specific implementation of step S05 is as follows: the main control unit retrieves data packets from the queue to be sent, uploads them to the shore via the Beidou communication device, and waits for confirmation information from the shore. If no confirmation is received within the response timeout period, the sending status flag of the data packet is kept as pending transmission, and the number of transmissions is recorded. When the communication link is restored, the retransmission scheduler reschedules the data packets in the queue according to the principles of alarm priority, recent priority, and sequential retransmission. It prioritizes data packets with alarm data type, then selects real-time data packets with the latest timestamp, and finally retransmits ordinary historical data packets in ascending order of timestamp. After each transmission, the sending status flag is updated based on the confirmation result from the shore.

[0058] The specific implementation of step S06 is as follows: the main control unit counts the number of consecutive no-response times for each sensor based on the response timeout duration and maximum retry count in the sensor task table. When the number of consecutive no-response times reaches the fault judgment threshold, the main control unit performs a power-off operation on the sensor through the branch load power supply control module, and after a preset delay, re-energizes the sensor and attempts to collect data again. If the sensor still does not respond after the retry, the sensor is placed in a fault isolation state, power supply to it is stopped, and the corresponding fault code is uploaded to the shore end through the Beidou communication device. After the shore end issues a control command, the main control unit verifies the command content. If the verification is successful, it performs parameter query, parameter modification, sensor reset, or historical data retransmission operations according to the command type. The fault judgment threshold is determined based on the normal communication frame error rate and allowable missed detection rate of each sensor; the delayed power-on time is not less than the power-on stabilization time specified in the corresponding equipment technical manual.

[0059] A second aspect of the present invention provides a computer-readable storage medium storing program instructions, which, when executed in a computer, are used to perform the above-described method for the electrical data acquisition and BeiDou communication fault-tolerant transmission of marine buoys.

[0060] A third aspect of the present invention provides a marine buoy electrical data acquisition and BeiDou communication fault-tolerant transmission system, comprising the aforementioned computer-readable storage medium. The system can be any one of a computer, a server, or a microcontroller. The computer-readable storage medium is disposed within the system, and the system is provided with a microprocessor that executes the program instructions stored in the computer-readable storage medium.

[0061] Specifically, the principle of this invention is as follows: The reason why the solution of this invention can solve the above-mentioned technical problems lies in transforming the scheduling problem and the prediction problem into continuous evolution processes with physical constraints, rather than relying on discrete rule judgments. At the scheduling level, the dynamic potential energy field model abstracts the urgency of each sensor acquisition task into a point mass, and the bus occupancy and power consumption limit into a potential energy field. By simplifying the point mass dynamic equation through real-time calculation, high-priority tasks naturally obtain bus access rights, while conflicting tasks are pushed into the waiting queue due to repulsion. This continuous field evolution mechanism naturally encodes the mutual exclusion relationship, eliminating the need for additional critical section protection logic and eliminating bus access conflicts at the physical level. At the prediction level, the energy functional physical constraint jump cycle prediction model distinguishes between stable discharge and violent charge and discharge conditions through a dynamic gating jump mechanism, and adopts two calculation paths: step jump and point-by-point fine iteration, balancing computational efficiency and prediction accuracy. At the same time, the electrochemical polarization equation and the thermodynamic monotonous degradation law are added as regularization terms to the loss function, so that the prediction results still maintain physical rationality on out-of-distribution samples. The outputs of the two models further drive adaptive sampling period adjustment and branch power supply control, forming a closed-loop control logic from data acquisition to energy management. This ensures that the system can simultaneously guarantee both the orderliness of bus access and the accuracy of energy status, thus solving the technical problem of the difficulty in balancing concurrent acquisition by multiple sensors and battery status assessment.

[0062] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0063] The specific implementation of step S01 is as follows: During the initialization phase, the main control unit establishes a sensor task table. This table is stored in the STM32's on-chip memory in the form of a structured array, and each entry contains a sensor number. Interface type, power supply channel number, sampling period Communication command content, response timeout duration Maximum number of retries Data parsing function pointer, upper limit of normal range Lower limit of measurement range and rate of change threshold ,in , The total number of sensors. The dynamic flow scheduling model for the kinetic potential energy field sensors runs in real time during timer interrupts, and its core is to maintain the virtual mass state for each acquisition task. Each sensor task in The combined force of moments The formula is expressed as follows:

[0064] ;

[0065] In the formula, For the first The virtual mass of each sensor task corresponds to a weight based on the urgency of data acquisition. The value ranges from 8 to 10 for alarm-type sensors, 6 to 8 for power status sensors, 4 to 6 for hydrological sensors, and 2 to 4 for meteorological sensors, rounded to the nearest integer. for Time of the first The virtual gravitational acceleration for each task increases as the acquisition deadline approaches, and is calculated linearly by the scheduler based on the remaining deadline time. The repulsive field coefficient is calibrated by a laboratory equivalent testing platform. and The first , The virtual charge of each task, and the conflict flag only when two tasks share the same RS485 bus. ,otherwise for Constant Task With the task The virtual distance on the scheduling timeline, i.e., the absolute value of the difference between the scheduled execution times of the two tasks. for The global damping coefficient at any given time, when the total power consumption of the system... Approaching the power consumption safety threshold Automatically increases For the first The virtual velocity of a task in the scheduling space represents the rate of progress of the task on the scheduling time axis. The third term... For damping term, , , Both the repulsive term and the quantity of repulsion are dimensionless scheduling priority quantities, and their dimensions are unified within the potential energy field model framework. Power consumption safety threshold. The instantaneous peak power consumption of each sensor when it is powered on simultaneously was measured using a laboratory equivalent test platform. Combined with the battery's maximum continuous discharge current and nominal voltage Determine using the following formula:

[0066] ;

[0067] In the formula, The power consumption safety factor is typically set between 0.75 and 0.85. The unit is , The unit is , The unit is The units on both sides of the equation are power. The task with the greatest combined force gets priority access to the bus and enters the execution queue, while conflicting tasks are pushed into the waiting queue, thus achieving lock-free bus access scheduling.

[0068] The specific implementation of step S02 is that the main control unit performs a three-level quality assessment on each frame of acquired data sequentially. The communication-level assessment includes frame header identifier matching and frame length field... Compared with the actual number of bytes Consistency verification, including calculating the cyclic redundancy check value for the entire frame and comparing it with the frame end check field. Physical quantity determination requires the parsed physical quantities... satisfy Otherwise, set the corresponding data quality flag to the physical quantity out-of-range state. The trend level determination formula is expressed as follows:

[0069] ;

[0070] In the formula, For the first The first sensor Second sample value, As the rate of change threshold, the maximum rate of change of each physical quantity under normal sea state is statistically analyzed based on historical observation data of the target sea area. ,according to Calculation, where The value is typically between 1.5 and 2.0, and is determined after on-site verification. and The dimensions of all are the same as the corresponding physical quantities The dimensions are consistent. The results of the Level 3 judgment are written into the data quality flag and fault code field, and packaged together with the collected data into a unified data packet.

[0071] The specific implementation of step S03 is that the energy functional physics constraint jump cycle prediction model uses the battery terminal voltage. Charging and discharging current and temperature The time-series window vector is composed of the following formula:

[0072] ;

[0073] In the formula, for 3D column vector, This is the length of the time window, typically ranging from 20 to 60 time steps. The feature embedding fully connected layer will... Mapping to latent space representation The formula is expressed as follows:

[0074] ;

[0075] In the formula, To embed the weight matrix, the dimension is , The latent space dimension typically ranges from 32 to 64. Here is the embedding layer bias vector, with dimension . For piecewise linear activation functions, the hyperbolic tangent function is approximated using shift-addition operations. for 2D latent space feature vectors; both sides of the equation are 2D latent space feature vectors; The dimensionless vector is represented by a dimensionless hidden layer representation. The core two-layer recurrent neural network encoder iteratively processes the temporal features, reaching the first hidden layer representation in the first layer. Step to determine the rate of change of terminal voltage The calculation formula is: ,in The sampling time interval is expressed in units of 1 / 2. , Units are ;like Then the step jump function is activated to pass the hidden state in a weighted residual manner, as expressed in the following formula:

[0076] ;

[0077] In the formula, The last valid hidden state vector of the main loop that is closest to the current step. The hidden state vector of the next effective main loop. Number of time steps skipped For step residual weight scalar The weight matrix for step-connection has dimensions of . This is the hidden state vector of the current step after the jump, with dimension . The dimensions on both sides of the equation are unified to the dimensionless hidden layer state. The threshold for determining stable discharge is expressed in units of... This is determined statistically from laboratory battery discharge characteristic test data. Then it jumps to the fine-tuning loop branch, using a single-layer recurrent neural network for high-precision point-by-point iteration, whose hidden state vector The dimension typically ranges from 16 to 24. The dual-branch outputs are concatenated into learnable weights at the fusion layer. and The weighted summation is expressed by the following formula:

[0078] ;

[0079] In the formula, The main loop branch is in the 1st... The output hidden state vector of the step To fine-tune the branch in the first The output hidden state vector of the step and For learnable scalar fusion weights The output vector after fusion is a dimensionless hidden layer vector on both sides of the equals sign, with unified dimensions. The capacity decay path selection submodule abstracts different charging and discharging history paths as a set of directed graph nodes. , The total number of nodes and the edge weights are: For path To path The rate of health degradation is expressed as the decrease in health status per cycle; the optimal degradation path that best matches the current battery's historical state is selected in real time during the inference phase using the Dijkstra algorithm. The slope of the decline path As an additional bias injected into the output layer, the formula is as follows:

[0080] ;

[0081] In the formula, For the remaining capacity of the battery With health status A two-dimensional column vector composed of predicted values The output layer weight matrix has dimensions of . , Output dimension for fusion layer The output layer bias vector has a dimension of 2. Scalar of the decay slope of the optimal path The bias direction vector, with a dimension of 2, is used to specify the fading bias at... and Distribution ratio of the two components , and All values ​​are dimensionless normalized predictions with uniform dimensions. The model's total loss function... The formula is expressed as follows:

[0082] ;

[0083] In the formula, The mean square error data fitting term and The physical constraint regularization weight coefficients have an initial value range of 0.1 to 0.3 and are dynamically adjusted by the model accuracy control function. As a constraint term in the electrochemical polarization equation, the terminal voltage needs to be predicted. Estimating terminal voltage using equivalent circuit model Deviations between these are penalized, as expressed in the following formula:

[0084] ;

[0085] In the formula, The total number of time steps within a single training batch For the model in the first The predicted terminal voltage of the step output, in units of The estimated terminal voltage is derived based on the equivalent circuit model, and the unit is 1. Units are The dimensions on both sides of the equal sign are consistent. As a monotonicity penalty item for health status, it requires The sequence is monotonically non-increasing in time, as expressed by the following formula:

[0086] ;

[0087] In the formula, For the first Predict health status values ​​step by step; if a local rebound occurs, apply additional penalties. , and After normalizing all three terms to the same order of magnitude, a weighted sum is taken to ensure that... Dimensionality is unified. The model accuracy control function is based on a comprehensive control index. Dynamic adjustment and , The mean square error of the current validation set Physical constraint violation rate Mean absolute error of battery remaining capacity prediction The result is obtained by weighted summation after normalization of the three terms; when At fixed step size Decrease and ,when Time remains unchanged, when At fixed step size Increase and The segment boundaries are not preset to be fixed at 0.3 and 0.7, but are determined by the Pareto front inflection point of the prediction accuracy and physical constraint satisfaction rate on the validation set. The main control unit is based on... and Combined with real-time terminal voltage Perform adaptive sampling period adjustment, high voltage threshold Low voltage threshold and minimum voltage threshold Calibration based on actual battery and load test conditions:

[0088] For a 12V rated sensor, 11V can continue to operate only if the device's allowed input range covers 11V and the power supply is stable; otherwise, the controller enters energy-saving or protection mode.

[0089] The high voltage threshold, low voltage threshold, and minimum voltage protection threshold are calibrated based on the battery chemistry, number of cells in series, remaining BMS capacity, minimum single-cell voltage, temperature, load current, and minimum operating voltage of critical loads, and a hysteresis range is set to avoid frequent switching near the threshold.

[0090] The specific implementation of steps S04 to S06 can be briefly described as follows. In step S04, the Allwinner V40 storage control unit maintains a log buffer in memory. When the amount of data in the buffer reaches an integer multiple of the logical block size and file system block size reported by the storage device, it performs aligned batch writing. The trigger quantity is determined by actual measurement of the target device. At the same time, it maintains a queue to be sent on the solid-state drive. Each record includes the data packet sequence number, collection timestamp, data type, priority, number of transmissions, last transmission time, transmission status flag, and cyclic redundancy check value fields. In step S05, the main control unit uploads the data packet via the Beidou communication device. If no confirmation is received from the shore end, it remains in the pending transmission state. After communication is restored, retransmission is scheduled according to the principle of alarm priority, recent priority, and sequential retransmission, and the transmission status flag is updated in sequence. In step S06, when the number of consecutive unresponsive times of the sensor reaches the fault judgment threshold (based on the normal communication frame error rate and the allowable missed alarm rate according to the sensor task calibration), the main control unit performs a circuit power-off, delayed power-on (the delay is not less than the power-on stabilization time specified in the corresponding sensor technical manual), and retry acquisition process for the sensor. If there is still no response after retry, it is set to the fault isolation state and the fault code is uploaded. After the shore end sends the instruction, the main control unit verifies the instruction and performs query, parameter modification, reset, or historical data retransmission operations.

[0091] To better understand and implement this invention, a specific application scenario, Example 2, is provided below: The method described in this invention is verified under simulated operating conditions on a joint debugging platform with the same hardware configuration as the target buoy. During the test, the buoy is equipped with twelve data acquisition nodes of four types: water temperature sensor, salinity sensor, wave height sensor, and meteorological sensor. The mass point weight assignments for each type of sensor in the sensor task table are set according to the urgency level of data acquisition, with alarm types having a weight of 9, power status types having a weight of 7, hydrological types having a weight of 5, and meteorological types having a weight of 3. For example... Figure 2As shown, the bus waiting time distribution of various sensor tasks under potential energy field scheduling was recorded during the joint debugging process. The results show that the average waiting time of high-weight tasks is significantly shorter than that of low-weight tasks. During the joint debugging process, abnormal sea state data was replayed, and the wave height sensor data showed a sudden trend change. The main control unit added a trend change abnormality flag bit to the corresponding frame based on the trend-level quality judgment, and simultaneously generated a fault code and uploaded it to the shore. The specific statistical results are shown in Table 1.

[0092] Table 1. Statistical Table of Abnormalities in Level 3 Quality Judgment

[0093]

[0094] During the test, under the conditions of an ambient temperature of 25±1℃, a 12V deep-cycle lead-acid battery, and a consistent load configuration, the battery completed one working charge-discharge cycle. The energy functional theory-constrained jump-cycle prediction model continuously outputs the battery's remaining capacity (SOC) and state of health (SOH) based on battery terminal voltage, charge / discharge current, and temperature time-series data. Four representative test points during the discharge phase were selected, and the model-predicted SOC was compared with the coulomb integral calibration SOC corrected by rated capacity testing. The results are shown in Table 2. The terminal voltage in the table is the open-circuit voltage measured after the battery has stopped charging and discharging and has been left to stand for at least 6 hours. It is only used to identify test points and is not directly used as the sole basis for SOC conversion. The battery SOH is calibrated by the ratio of the actual usable capacity after multiple charge-discharge cycles to the initial rated capacity, without verification using a single charge-discharge cycle or a single terminal voltage.

[0095] Table 2 Comparison of Battery Status Prediction

[0096]

[0097] like Figure 3As shown (the fixed voltage values ​​in the figure are not used as a unified control threshold for different battery systems or rated 12V sensors; that is, the voltage values ​​in the figure are only used to illustrate the hierarchical relationship, and the actual threshold is calibrated according to the specific battery system, BMS parameters, and minimum operating voltage of the load), the four sampling modes divided by the main control unit based on the battery terminal voltage during the joint debugging process are compared with the actual branch power supply status. The results show that the system can automatically shut down the power supply to meteorological sensors when the terminal voltage drops below the low voltage threshold, while only alarm and hydrological sensors continue to collect data. During the joint debugging process, a communication interruption was constructed through link simulation. During the interruption, the number of data packets in the queue to be sent continued to accumulate. After communication was restored, the retransmission scheduler retransmitted the packets one by one according to the shore-end confirmation results based on the principles of alarm priority, recent priority, and sequential retransmission. Alarm data packets entered the sending queue first, and their completion time was recorded in the joint debugging log. During the write process during the test, the storage control unit merges the originally scattered multiple small data writes into batch writes through the incremental merging of the memory log buffer and the sector alignment write mechanism. The program records the number of storage writes and compares them with the frame-by-frame write method; the specific reduction is based on the actual test results of the target memory. Figure 4 This is a schematic diagram of the hardware system involved in this embodiment, showing the electrical connections between the STM32 microcontroller main control unit, the storage control unit, the power management and power supply control module, the Beidou communication unit, and the multi-source sensor array.

[0098] The advancements of this invention compared to traditional methods are as follows: Traditional fixed-priority polling methods are prone to bus access conflicts when multiple sensor tasks arrive simultaneously, while this invention uses a potential energy field evolution mechanism to naturally stagger the access times of conflicting tasks, fundamentally reducing the possibility of bus conflicts; Traditional fixed-threshold judgment methods are prone to distorted estimations when the battery is in an unsteady state, while this invention uses a physical constraint regularization term to ensure that the prediction results always conform to the electrochemical and thermodynamic evolution laws, fundamentally improving the reliability of energy state assessment under complex operating conditions; Traditional frame-by-frame writing methods cause frequent erasures and writes to the storage medium, while this invention uses incremental merging and sector-aligned writing mechanisms to fundamentally reduce the physical erasure and write frequency.

[0099] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for fault-tolerant transmission of marine buoy electrical data acquisition and BeiDou communication, characterized in that, Includes the following steps: A sensor task table is established for each sensor. Based on the dynamic flow scheduling model of time-sharing polling for dynamic potential energy field sensors, each acquisition task is scheduled, driving the main control unit to execute the time-sharing polling acquisition of each sensor in sequence according to time slices. The main control unit performs three levels of quality judgment on the collected data: communication level, physical quantity level, and trend level. It adds data quality flag bits and fault codes to each frame of data and packages the data into a unified data packet. The energy functional physics constraint jump cycle prediction model is based on real-time collected battery terminal voltage, charging and discharging current and temperature, and outputs the remaining battery capacity and battery health status. The main control unit performs adaptive sampling period adjustment and branch load power supply control according to the remaining battery capacity and battery health status. The main control unit sends data packets to the storage control unit, which writes the data to the solid-state drive through incremental merging of the memory log buffer and sector-aligned writing mechanism, and maintains the queue to be sent and the sending status. The main control unit uploads the data packet to the shore end through the Beidou communication device. If no confirmation is received from the shore end, the data packet is kept in the state of waiting to be sent. After communication is restored, the cached resending is performed according to the principles of alarm priority, recent priority, and sequential resending. When the number of consecutive unresponsive times of the sensor reaches the fault judgment threshold, the main control unit executes the process of power-off, delayed power-on, and retry data acquisition. If there is still no response after the retry, the sensor is placed in a fault isolation state and a fault code is uploaded to the shore. After the shore sends a command, the main control unit verifies the command and performs query, parameter modification, reset, or historical data retransmission operations.

2. The method according to claim 1, characterized in that, The main control unit is an STM32 microcontroller.

3. The method according to claim 1, characterized in that, The sensor task table includes sensor number, interface type, power supply channel number, sampling period, communication command content, response timeout duration, maximum number of retries, data parsing function pointer, normal range upper and lower limits, and rate of change threshold.

4. The method according to claim 1, characterized in that, The dynamic flow scheduling model of the dynamic potential energy field sensor in time-sharing polling abstracts the urgency of acquisition as the mass of a point mass, and abstracts interface conflicts and power consumption limits as a virtual resistance potential energy field.

5. The method according to claim 4, characterized in that, Conflicting tasks on the shared bus are pushed into the waiting queue due to virtual charge repulsion, and the potential field damping coefficient increases when the total power consumption of the system approaches the safety threshold.

6. The method according to claim 1, characterized in that, The communication-level quality assessment specifically verifies the consistency between the frame header identifier, the frame length field and the actual data length, and the cyclic redundancy check value.

7. The method according to claim 1, characterized in that, The physical quantity quality determination and the trend level quality determination are respectively range comparison and adjacent sampling difference comparison.

8. The method according to claim 1, characterized in that, The energy functional physical constraint jump cycle prediction model is based on a multi-layer recurrent neural network structure, which includes a feature embedding fully connected layer and a two-layer recurrent neural network encoder.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions, which, when executed in a computer, are used to perform the marine buoy electrical data acquisition and BeiDou communication fault-tolerant transmission method according to any one of claims 1-8.

10. A marine buoy electrical data acquisition and BeiDou communication fault-tolerant transmission system, characterized in that, The system comprises the computer-readable storage medium of claim 9, wherein the system is a computer, the computer-readable storage medium is disposed within the system, and the system is provided with a microprocessor that executes program instructions stored in the computer-readable storage medium.

Images