Fishery water area multi-parameter intelligent monitoring system

Abstract

The invention relates to the technical field of intelligent sensors, and particularly discloses a fishery water area multi-parameter intelligent monitoring system. The system comprises intelligent monitoring nodes, a data aggregation gateway and a central processing platform. The node core is an event-driven self-adaptive sleep wake-up controller, the event occurrence probability of water quality sudden abnormity is deduced and calculated based on the multi-parameter joint probability, a dynamic energy budget coefficient is coupled, the duration of a next sleep period is determined in a self-adaptive mode through a decision function, and dynamic balance of energy consumption constraint and event capture sensitivity is achieved. Through'end-edge-cloud 'cooperation, the system can improve the capture rate and early warning timeliness of sudden abnormal water quality events.

Description

Technical Field

[0001] This invention belongs to the field of intelligent sensor technology, and specifically relates to a multi-parameter intelligent monitoring system for fishery waters. Background Technology

[0002] In the fields of aquaculture and fishery resource management, real-time and accurate monitoring of the aquatic environment is a key technological support for ensuring the quality of aquatic products, optimizing aquaculture efficiency, and providing early warning of ecological risks. The integrated application of Internet of Things (IoT) technology and intelligent sensing devices provides the foundation for building an automated monitoring network. Multi-parameter sensor nodes deployed in water bodies are the basic units for collecting core water quality data such as dissolved oxygen, pH, temperature, and ammonia nitrogen.

[0003] Fisheries water monitoring systems based on wireless sensor networks are an important technological direction. This system aims to periodically collect water quality parameters through distributed deployment of intelligent sensor nodes, and transmit the data via wireless network to the cloud or local server for analysis and processing, thereby achieving unmanned and intelligent monitoring of large areas of water.

[0004] Existing technologies typically employ a periodic sleep and wake-up mechanism to extend the battery life of sensor nodes. However, this fixed or simple threshold-based sleep strategy has drawbacks: during low-power sleep periods, sensor nodes are inactive and unable to detect sudden changes in the external environment. When water quality parameters deteriorate sharply within the sleep interval, existing systems may completely miss monitoring this critical event due to the node's failure to wake up in time, resulting in a high rate of loss of critical water quality anomaly data.

[0005] This data gap not only reduces the reliability and timeliness of monitoring systems, but also risks causing significant economic losses or ecological accidents in fisheries due to the failure to promptly detect emergencies such as oxygen depletion or pollution leaks. Therefore, ensuring that intelligent monitoring nodes have a sensitive response capability to sudden water quality events without significantly increasing energy consumption has become an urgent technical challenge. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-parameter intelligent monitoring system for fishery waters to solve the problems of sensor nodes being unable to respond to sudden water quality changes and the high rate of key data loss caused by the fixed dormancy strategy in the prior art.

[0007] This invention provides a multi-parameter intelligent monitoring system for fishery waters, which includes multiple intelligent monitoring nodes deployed in the target waters, at least one data aggregation gateway, and a central processing platform in the cloud or locally.

[0008] The intelligent monitoring nodes are responsible for collecting and processing water quality parameter data, the data aggregation gateway is responsible for data relay and aggregation of nodes within the region, and the central processing platform is responsible for global data analysis, model training and command issuance.

[0009] The intelligent monitoring node includes a sensor array, a microcontroller unit, a wireless communication module, an energy management module, and an event-driven adaptive sleep / wake-up controller.

[0010] The sensor array includes at least a dissolved oxygen sensor, a pH sensor, a temperature sensor, and an ammonia nitrogen sensor.

[0011] Used for synchronous acquisition of multi-dimensional water quality parameters. The microcontroller unit has built-in signal conditioning circuitry and an analog-to-digital converter, which is responsible for filtering, amplifying, and digitizing the raw analog signals output by the sensor array.

[0012] The event-driven adaptive sleep / wake controller is the core logic unit of the intelligent monitoring node, and its operation does not depend on a fixed time period or a simple single-parameter threshold judgment.

[0013] The controller executes a hibernation decision algorithm based on multi-parameter joint probabilistic inference and dynamic energy budget.

[0014] The algorithm is executed as follows:

[0015] The microcontroller unit wakes up at each preset basic sampling cycle, drives the sensor array to complete a full parameter acquisition, and sends the acquired water quality parameter dataset to the event-driven adaptive sleep-wake controller.

[0016] The controller first compares the current dataset with the historical normal operating condition parameter model stored in local memory.

[0017] The historical normal operating condition parameter model is established by the central processing platform through analysis of long-term monitoring data and periodically distributed to each node. This model defines the probability distribution range of each water quality parameter and the coupling relationship between them under steady state.

[0018] The comparison process is as follows: the controller calculates the normalized distance of each parameter value in the current dataset from the center value of its historical normal operating condition probability distribution interval, and calculates the comprehensive deviation index based on the preset weight coefficient.

[0019] At the same time, the controller calculates the relationship between each parameter within the current parameter set and the matching degree between it and the parameter coupling relationship defined in the historical model to obtain the relationship consistency index.

[0020] The comprehensive deviation index and the relationship consistency index together constitute a multidimensional feature vector.

[0021] The controller inputs the multidimensional feature vector into a localized lightweight anomaly classifier.

[0022] The anomaly classifier is a machine learning model pre-trained and distributed by a central processing platform. Its input is the multidimensional feature vector, and its output is the probability value that the water quality status at the current sampling time belongs to a "sudden anomaly" event. This probability value is denoted as the event occurrence probability. The event occurrence probability is a value between 0 and 1.

[0023] The energy management module monitors the remaining battery power of the nodes in real time and calculates the dynamic energy budget coefficient. The dynamic energy budget coefficient is positively correlated with the remaining battery power, and its specific calculation method is as follows: the ratio of the current remaining battery power to the total battery capacity is mapped through a preset smoothing function.

[0024] The event-driven adaptive sleep / wake controller ultimately calculates the duration of the next sleep cycle based on the probability of the event occurrence and the dynamic energy budget coefficient through a determined decision function.

[0025] The logic of the decision function is: the higher the probability of an event occurring, the shorter the calculated duration of the next sleep cycle;

[0026] The lower the dynamic energy budget coefficient, the longer the calculated duration of the next hibernation cycle, but there is an upper limit set by the probability of the event occurring.

[0027] Specifically, the decision function ensures that when the probability of an event occurs is greater than a preset high probability threshold, the next sleep cycle will be forced to be an extremely short active monitoring cycle, regardless of the dynamic energy budget coefficient, so that the node enters a high-frequency sampling mode.

[0028] Conversely, when the probability of an event occurring is less than the preset low probability threshold, the sleep duration is adjusted within a relatively wide normal range based on the dynamic energy budget coefficient to achieve energy saving.

[0029] Through this decision-making mechanism, the system can achieve a dynamic optimal balance between energy consumption constraints and event capture sensitivity.

[0030] Furthermore, the training and update mechanism of the localized lightweight anomaly classifier is as follows:

[0031] The central processing platform continuously collects labeled data uploaded by all intelligent monitoring nodes. The labeled data includes multi-dimensional feature vectors and final event labels confirmed by the platform's expert system or by manual review.

[0032] The central processing platform uses this accumulated labeled data to periodically retrain the anomaly classifier model.

[0033] When the performance evaluation metrics of the new model are better than those of the old model currently deployed on the node, the central processing platform packages the parameters of the new model and the corresponding historical positive operating condition parameters, and distributes them to each intelligent monitoring node through the data aggregation gateway in a differential update manner.

[0034] After receiving the update package, the intelligent monitoring node, under the control of the microcontroller unit, completes the replacement and verification of model parameters during idle periods, thereby realizing the continuous evolution of the classifier's capabilities.

[0035] Furthermore, in addition to being responsible for data forwarding, the data aggregation gateway also has edge collaborative computing capabilities.

[0036] When a smart monitoring node enters a high-frequency sampling mode due to a persistently high probability of an event, it will send a collaborative computing request flag along with the data packet it uploads to the gateway. Upon receiving this flag, the data aggregation gateway will proactively retrieve recent historical data from the requesting node's neighboring nodes.

[0037] The gateway has a built-in edge computing module that runs a spatial correlation analysis algorithm to determine whether a sudden abnormal event is a local event or a regional event with a spatial diffusion trend.

[0038] If the analysis results indicate that the anomaly has spatial diffusion, the data aggregation gateway will no longer passively wait for instructions from the central processing platform, but will directly broadcast temporary warning instructions to all relevant nodes within its communication range. The instructions will include a parameter suggesting a shortened sleep period.

[0039] Upon receiving this instruction, the surrounding nodes will temporarily adopt the parameters suggested by the gateway in their event-driven adaptive sleep / wake-up controllers, thereby entering a higher level of monitoring in advance and achieving proactive defense against the spread of potential risks.

[0040] Furthermore, the central processing platform includes a data warehouse, a model training engine, an expert knowledge base, and a visual early warning interface.

[0041] A data warehouse is used to store all historical and real-time data.

[0042] The model training engine is responsible for training and optimizing the aforementioned anomaly classifier model.

[0043] The expert knowledge base stores the safety range requirements of various water quality parameters for different fishery species at different growth stages, as well as the parameter change patterns of various typical pollution or ecological events.

[0044] The visual early warning interface comprehensively displays monitoring data, node status, event probability, and early warning information.

[0045] One of the core functions of the central processing platform is to perform cross-node trend risk assessment, the process of which is as follows:

[0046] The platform periodically analyzes the data of all nodes, identifies parameter change patterns that are spatially clustered and temporally trending, and matches these patterns with event patterns in the expert knowledge base.

[0047] When the matching degree exceeds the set threshold, the platform determines that there is a systemic risk and issues a global monitoring strategy enhancement instruction to all nodes in the affected area through the data aggregation gateway, uniformly adjusting the decision function parameters of the relevant nodes to cope with possible large-scale water quality events.

[0048] Furthermore, the sensor array of the intelligent monitoring node adopts a modular packaging and interface design. The microcontroller unit connects to each sensor module through a standardized digital interface.

[0049] Each sensor module integrates a self-diagnostic circuit that can periodically test the sensor probe's sensitivity, zero-point drift, and response time.

[0050] The self-diagnostic results will be uploaded as metadata attached to the regular water quality data package.

[0051] The central processing platform is equipped with a sensor health assessment module, which analyzes the rationality of long-term uploaded self-diagnostic data and water quality parameter data to determine whether the sensor has experienced performance degradation or malfunction.

[0052] When the assessment results show that the health status of a certain sensor is less than the preset threshold, the platform will generate a maintenance alarm and highlight the location of the node and the type of sensor suspected of being faulty in the visualization interface, guiding maintenance personnel to perform accurate maintenance or replacement.

[0053] Furthermore, the communication protocol of the system employs a mechanism combining time division multiple access and frequency hopping to enhance reliability.

[0054] The data aggregation gateway is responsible for allocating communication time slots to all intelligent monitoring nodes within its jurisdiction.

[0055] When an event-driven adaptive sleep / wake controller decides to wake up and send data, it must first synchronize time slots with the gateway.

[0056] When the channel is interfered with, the wireless communication module can automatically switch to a preset backup frequency point for communication according to the protocol.

[0057] For emergency data packets marked as having a high probability of occurrence, the system assigns them the highest communication priority, allowing them to access the channel in a contention-based manner in non-designated time slots and enjoy redundant transmission guarantees, ensuring that critical alarm information can be successfully delivered with the highest probability.

[0058] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0059] 1. This invention fundamentally changes the traditional fixed-period sleep mode by introducing an event-driven adaptive sleep / wake controller and its core algorithm. Based on multi-parameter joint probability inference, the system calculates the probability of sudden water quality anomalies in real time and couples this with dynamic energy budgeting for sleep decisions, enabling monitoring nodes to intelligently switch between low-power normal operation and high-response emergency state. This essentially solves the blind spot problem in monitoring sudden water quality events, increasing the detection rate of key anomalies to nearly 100% without significantly increasing the system's average energy consumption, greatly enhancing the reliability and timeliness of the monitoring system.

[0060] 2. This invention constructs an intelligent architecture that integrates "end-edge-cloud" collaboration. A lightweight local classifier is deployed at the terminal node for rapid initial assessment; spatial correlation analysis is introduced at the edge gateway to achieve proactive collaboration and early warning of localized risks; and global model training, trend analysis, and strategy optimization are performed on the central platform in the cloud. This layered collaborative mechanism not only reduces the data processing pressure and communication bandwidth consumption in the cloud but also improves the system's overall response speed and decision-making foresight for regional emergencies through real-time collaboration at the edge, forming a three-dimensional monitoring capability from single-point perception to regional joint defense.

[0061] 3. This invention achieves a closed loop of self-sensing, self-optimization, and self-maintenance for the monitoring system. By continuously collecting data and retraining the model, the system's anomaly classification capabilities continuously evolve with changes in the environment and data distribution. Simultaneously, by combining sensor self-diagnosis with platform health assessment, real-time monitoring and predictive maintenance of the monitoring equipment's own status are achieved. This ensures the accuracy and stability of the entire monitoring network's long-term operation, reduces maintenance costs, and enables the system to adapt to complex and ever-changing real-world fishery environments, possessing advantages such as long lifespan and high availability. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the overall technical architecture of the intelligent multi-parameter monitoring system for fishery waters proposed in this invention;

[0063] Figure 2 This is a schematic diagram of the core principle framework of the event-driven adaptive sleep / wake controller in this invention;

[0064] Figure 3 This is a flowchart illustrating the logical process of multi-parameter joint probability inference and dormancy decision-making by the intelligent monitoring node in this invention.

[0065] Figure 4This is a schematic diagram of the multi-level interaction relationship and data flow of the data aggregation gateway performing edge collaborative computing and spatial early warning in this invention;

[0066] Figure 5 This is a closed-loop process framework diagram of model training and updating and global trend judgment under the "end-edge-cloud" collaborative architecture in this invention. Detailed Implementation

[0067] This invention provides a multi-parameter intelligent monitoring system for fishery waters. Please refer to the appendix. Figures 1 to 5 The overall architecture of the system includes multiple intelligent monitoring nodes deployed in the target water area, at least one data aggregation gateway, and a central processing platform in the cloud or locally.

[0068] The intelligent monitoring node is responsible for collecting and processing water quality parameter data at its location.

[0069] The data aggregation gateway is responsible for the data relay and aggregation of all intelligent monitoring nodes within its wireless communication coverage area.

[0070] The central processing platform is responsible for storing global data, performing in-depth analysis, training and optimizing machine learning models, and issuing monitoring strategy instructions.

[0071] The three work together to form a complete three-layer intelligent monitoring network of "end-edge-cloud".

[0072] Intelligent monitoring nodes are the basic sensing units of the system. Please refer to the attached document. Figure 1 and attached Figure 2 Each intelligent monitoring node physically includes a sensor array, a microcontroller unit, a wireless communication module, an energy management module, and an event-driven adaptive sleep / wake-up controller.

[0073] These components are integrated into a waterproof, corrosion-resistant housing and are electrically connected and exchange signals via an internal circuit board.

[0074] Sensor arrays are components that allow nodes to interact directly with their environment.

[0075] The array contains at least four core water quality parameter sensors: dissolved oxygen sensor, pH sensor, temperature sensor, and ammonia nitrogen sensor.

[0076] These sensors employ modular packaging and standardized digital interface design.

[0077] The microcontroller unit establishes communication connections with each sensor module through a universal asynchronous transceiver interface or an integrated circuit bus interface.

[0078] Each sensor module integrates a self-diagnostic circuit.

[0079] This self-diagnostic circuit can be automatically activated and execute a series of diagnostic procedures periodically, such as every 24 hours or after every 100 valid measurements.

[0080] Diagnostic procedures include, but are not limited to:

[0081] A standard test signal is applied to the sensor probe to detect its sensitivity and response curve; the output value is measured in a targetless environment to evaluate zero-point drift;

[0082] Measure the response time of the sensor to a step signal.

[0083] The self-diagnostic circuit encodes the above diagnostic results, including sensitivity coefficient, zero-point offset, response time value, and diagnostic timestamp, into a metadata data packet in a specific format.

[0084] After each routine water quality data collection is completed, this metadata package will be packaged together with the water quality parameter data as additional information, ready for uploading.

[0085] The microcontroller unit is the core of the node's computing and control.

[0086] It integrates signal conditioning circuitry and a high-precision analog-to-digital converter.

[0087] The signal conditioning circuit receives raw analog voltage or current signals from each sensor in the sensor array.

[0088] This circuit includes a multi-stage programmable gain amplifier and an anti-aliasing filter. The programmable gain amplifier linearly amplifies weak signals according to preset sensor range and output characteristics, making its amplitude match the input range of the analog-to-digital converter.

[0089] The anti-aliasing filter adopts a Butterworth low-pass filter design, and its cutoff frequency is set according to the highest effective signal frequency of each water quality parameter. For example, the cutoff frequency of the temperature signal is 1 Hz, and the cutoff frequency of the dissolved oxygen signal is 0.1 Hz, which is used to filter out high-frequency noise and power frequency interference.

[0090] The conditioned analog signal is then fed into an analog-to-digital converter (ADC). The ADC uses at least 16 bits of resolution to digitize the signal in each channel at a sampling rate of 1000 times per second.

[0091] The microcontroller unit's firmware performs software filtering on the converted digital sequence, using algorithms such as moving average filtering or median filtering to further smooth the data. Finally, each sensor channel outputs a calibrated physical quantity representing the current water quality parameter at a single sampling moment, in units of milligrams per liter (mg / L), dimensionless, degrees Celsius, and milligrams per liter, respectively. These values ​​collectively constitute the water quality parameter dataset for the current moment.

[0092] The energy management module is responsible for the node's energy supply and status monitoring. At its core is a rechargeable lithium battery pack with a nominal voltage of 3.7 volts and a typical capacity of 10,000 mAh. The energy management module integrates a battery power metering chip and a DC-DC voltage conversion circuit.

[0093] The battery power metering chip monitors the battery's voltage, current, and temperature, and uses a combination of coulomb counting and voltage lookup table methods to calculate and report the battery's remaining power in real time. This value is in milliampere-hours and has an accuracy of ±1%.

[0094] The DC-DC voltage conversion circuit converts the battery voltage into various stable operating voltages required by the sensor array, microcontroller unit, and wireless communication module, such as 3.3 volts and 5.0 volts. The energy management module periodically reports status parameters such as the remaining battery charge, current output current, and battery temperature to the microcontroller unit via the integrated circuit bus interface, for example, every 60 seconds.

[0095] The wireless communication module is responsible for bidirectional wireless data transmission between the node and the data aggregation gateway.

[0096] This module employs a long-range wireless RF chip operating in the 470 MHz to 510 MHz frequency band, using Gaussian frequency shift keying modulation. The transmit power is software-adjustable between 10 milliwatts and 100 milliwatts. The communication protocol combines time division multiple access with frequency hopping to form a self-organizing network.

[0097] Specifically, the data aggregation gateway acts as a network coordinator, assigning a unique network address and a fixed communication time slot to each intelligent monitoring node within its jurisdiction.

[0098] The time slot period is 10 seconds, and each time slot lasts for 100 milliseconds. During non-communication periods, the wireless communication module is placed in deep sleep mode to save power. When a node needs to send data, the event-driven adaptive sleep / wake controller first instructs the wireless communication module to wake up and synchronize the time slot with the beacon frame sent by the gateway, and then completes the transmission of data packets within the specified time slot.

[0099] The communication data packets adopt a custom binary protocol format, which includes a frame header, source node address, destination gateway address, data packet type, data payload length, water quality parameter data or status information, cyclic redundancy check code, and frame trailer.

[0100] For a standard data packet, the payload includes timestamps, dissolved oxygen levels, pH levels, temperature, ammonia nitrogen levels, and sensor self-diagnostic metadata.

[0101] When the channel is subjected to persistent narrowband interference, causing the bit error rate to exceed 1 / 1000 for three consecutive times, the driver firmware of the wireless communication module will automatically switch the communication carrier frequency to a backup clean frequency point according to the preset frequency hopping sequence, such as switching from 490 MHz to 500 MHz, and notify the gateway in the next communication cycle to maintain link reliability.

[0102] The event-driven adaptive sleep / wake-up controller is the core logic unit of the intelligent monitoring node, running as a software algorithm on the microcontroller unit. This controller completely abandons the fixed-time-period sleep mode, relying instead on a sleep decision-making algorithm based on multi-parameter joint probabilistic inference and dynamic energy budgeting. Please refer to the attached... Figure 2 and attached Figure 3 The complete execution flow of this algorithm is as follows.

[0103] First, the microcontroller unit is woken up at each preset base sampling period. The base sampling period is a configurable parameter with a default value of 300 seconds.

[0104] After being woken up, the microcontroller unit drives the sensor array to complete a full parameter acquisition, that is, to simultaneously acquire the values ​​of four parameters: dissolved oxygen, pH, temperature, and ammonia nitrogen, forming the current water quality parameter dataset.

[0105] The dataset is immediately sent to the input buffer of the event-driven adaptive sleep / wake controller.

[0106] After receiving the new dataset, the controller initiates a multi-parameter joint probability inference process. The first step of this process is to compare the data with historical normal operating condition parameter models.

[0107] The historical normal operating condition parameter model is established by the central processing platform by analyzing all water quality data collected by the node in a stable state over a long period of time, such as the past 30 days, and is periodically sent to the node and stored in the flash memory of the microcontroller unit.

[0108] This model is not a simple threshold range, but a set of data structures that describe the statistical characteristics of each parameter.

[0109] For each water quality parameter, the model records its probability distribution characteristics under steady-state conditions, including the mean, standard deviation, and the normal range defined by 5% and 95%.

[0110] In addition, the model defines a coupling relationship matrix between parameters, which describes the typical range of correlation coefficients between the changing trends of any two parameters under normal operating conditions. For example, temperature and dissolved oxygen are usually negatively correlated, with a typical correlation coefficient range of -0.7 to -0.3.

[0111] The comparison process is specified by the calculation of two indicators. The first indicator is the comprehensive deviation index.

[0112] The controller calculates the normalized distance of each parameter value in the current dataset from the center value of its historical normal operating condition probability distribution interval.

[0113] For parameters Its current value is The historical average is The historical standard deviation is Then the normalized distance The calculation method is as follows Subsequently, the controller determines the weighting coefficients based on preset weighting factors. The deviations of each parameter are weighted and summed. The weighting coefficients are preset based on the importance and sensitivity of each parameter to fishery organisms. For example, dissolved oxygen has a weight of 0.4, ammonia nitrogen has a weight of 0.3, pH has a weight of 0.2, and temperature has a weight of 0.1.

[0114] Comprehensive Deviation Index The calculation formula is:

[0115] ;

[0116] This formula calculates the weighted degree of deviation of the current set of water quality parameters from the historical normal central level. The higher the value, the higher the probability of overall water quality abnormality.

[0117] The second metric is the relationship consistency metric. The controller calculates the degree of matching between the relationships between parameters within the current parameter set and the parameter coupling relationships defined in the historical model.

[0118] Specifically, the controller extracts short-term trend vectors from the current dataset, for example, by calculating the slope sign of each parameter using the three most recent sampled values.

[0119] Then, calculate the product of the trend signs of any two parameters in the current trend vector and compare it with the expected value of the correlation coefficient of the corresponding parameter pair in the historical coupling matrix.

[0120] The degree of matching is quantified by the proportion of parameter pairs with consistent statistical signs to the total number of parameter pairs.

[0121] For example, if historical models show a negative correlation between temperature and dissolved oxygen, while current data show that both are rising simultaneously, this pair of parameters is considered mismatched.

[0122] The final relationship consistency index It is a value between 0 and 1, where 1 indicates that it fully conforms to the historical coupling relationship and 0 indicates that it completely deviates from it.

[0123] Comprehensive Deviation Index Consistency index of relationship Together they form a two-dimensional feature vector The controller then inputs this feature vector into a localized, lightweight anomaly classifier.

[0124] The anomaly classifier is a machine learning model pre-trained and distributed by a central processing platform. It is implemented in the form of a decision tree or a micro neural network, and the model parameters are stored in the node flash memory.

[0125] The input to the classifier is a two-dimensional feature vector. Its output is a scalar value between 0 and 1, which represents the probability that the water quality status at the current sampling time belongs to a "sudden anomaly" event, denoted as the event occurrence probability. .

[0126] Meanwhile, the battery remaining power information provided in real time by the energy management module is used by the controller to calculate the dynamic energy budget factor. Let the current remaining battery power be... The unit is milliampere-hours (mAh), and the total battery capacity is... Then the power ratio Dynamic energy budget coefficient The function is obtained by mapping through a preset smoothing function, which is designed to encourage more active monitoring when the battery is full and to force extended hibernation to ensure survival when the battery is low.

[0127] A specific example of a mapping function is: when When ≥0.5, ;when When it is between 0.2 and 0.5, ;when When less than 0.2, . The value ranges from 0.5 to 1.0 and is positively correlated with the remaining battery power.

[0128] Event-driven adaptive sleep / wake controllers ultimately rely on the probability of event occurrence. With dynamic energy budget coefficient The duration of the next sleep cycle is calculated using a defined decision function. The core logic of the decision function is encapsulated in a lookup table or a piecewise linear function. Its basic principle is: The higher, the better The shorter; The lower, the better The longer, but It possesses the capability to enforce intervention. The specific decision-making rules are as follows:

[0129] First, set two probability thresholds: a high probability threshold and a low probability threshold. The value is 0.8, a low probability threshold. It is 0.2.

[0130] Second, if Then regardless Why is this value set by default? for This refers to the active monitoring cycle, typically 30 seconds. The node enters a high-frequency sampling mode.

[0131] Third, if The main basis is Decide Set the base sleep duration. The typical value is 600 seconds. Actual... For example, when hour, seconds; when hour, Seconds. However, the system has a maximum sleep duration limit. For example, 1800 seconds, to prevent the device from completely losing its monitoring capabilities due to prolonged sleep at extremely low battery levels.

[0132] Fourth, if Between and Between, Depend on and This is jointly determined. Through a linear interpolation function, it is made possible... from The corresponding value (by) (Decision) Smooth transition to corresponding .

[0133] The formula is expressed as:

[0134] ;

[0135] This formula ensures that, under moderate anomaly probabilities, the dormancy duration decreases linearly with increasing probability, while still being regulated by the energy budget.

[0136] Calculated Afterwards, the controller instructs the microcontroller unit to shut down the sensor array, wireless communication module, and most peripheral circuits, in addition to maintaining the real-time clock, and enter a low-power sleep mode.

[0137] The timer is set to T_sleep The microcontroller unit is woken up again after milliseconds to start the next working cycle.

[0138] After a node completes data collection and local processing, it needs to send the data to the data aggregation gateway.

[0139] The timing and priority of data transmission are managed by the event-driven adaptive sleep / wake controller in conjunction with the wireless communication module.

[0140] For regular data packets, the node sends them in the next available pre-allocated time slot. However, for packets marked as having a high probability of occurrence, i.e. ≥ The system assigns the highest communication priority to the data packets.

[0141] These types of packets are marked as urgent packets.

[0142] Upon generating an emergency data packet, the node will immediately interrupt its current sleep or waiting state and forcibly wake up the wireless communication module. The wireless communication module will first attempt to compete for channel access using the carrier sense multiple access conflict avoidance mechanism in a non-designated time slot.

[0143] If the channel is idle, send the data packet immediately.

[0144] Meanwhile, to ensure the reliable delivery of critical alarm information, emergency data packets enjoy redundant transmission protection.

[0145] After sending the first data packet, the sending end will wait to receive an acknowledgment frame from the gateway.

[0146] If no acknowledgment is received within 200 milliseconds, the sender will immediately retransmit after a random backoff, with a maximum of 3 retransmissions.

[0147] The data aggregation gateway also has the highest priority in processing urgent data packets, ensuring that they are received and forwarded quickly.

[0148] The data aggregation gateway acts as a bridge connecting intelligent monitoring nodes and the central processing platform. Please refer to the attached document. Figure 1 and attached Figure 4 The gateway hardware includes a high-performance processor, large-capacity storage, a long-range wireless communication module, and a backhaul communication module. The long-range wireless communication module is used to communicate with intelligent monitoring nodes within the jurisdiction, and its protocol is consistent with that of the nodes. The backhaul communication module typically uses a 4G or 5G cellular network module to establish a wide area network connection with the cloud-based central processing platform.

[0149] One of the core functions of a gateway is data relay and aggregation. The gateway continuously monitors its assigned time slots, receiving data packets from various nodes. After parsing and verifying the cyclic redundancy check (CRC) code, the received data packets are temporarily stored in a local buffer. According to a preset aggregation period, such as every 5 minutes, or when the buffer data volume reaches a certain threshold, the gateway packages data from multiple nodes into a batch and uploads it to the central processing platform via the backhaul network. The data packet contains the complete node address, timestamp, water quality parameters, self-diagnostic metadata, and the event occurrence probability value calculated by the node.

[0150] In addition to basic data forwarding, the data aggregation gateway possesses important edge collaborative computing capabilities. This capability is implemented by the edge computing module running on the gateway processor.

[0151] When a gateway receives a data packet from a smart monitoring node and parses it to find that the packet includes a collaborative computing request flag, this flag is typically set by the node in... If the value is continuously greater than a certain threshold, such as 0.6, and exceeds this threshold for three consecutive cycles, an automatic appending is triggered, and the edge collaborative computing process is activated.

[0152] The edge computing module first extracts the identity and latest water quality data of the source node that issued the request. Then, the module retrieves recent historical data from a locally stored historical data cache for 3 to 5 other geographically adjacent nodes, typically within a 500-meter radius. This historical data includes water quality parameter values ​​and event probability sequences for each node over the past hour.

[0153] The gateway's built-in spatial correlation analysis algorithm has started running.

[0154] The algorithm first performs time series alignment and preprocessing on the same water quality parameter, such as dissolved oxygen, for the source node and its surrounding nodes. Then, the algorithm calculates two core indicators:

[0155] The first is the spatial variation coefficient, which measures the degree of dispersion of this parameter value at each node at the current moment;

[0156] Second, the time lag correlation coefficient is used to analyze the order and correlation strength of outlier parameter values ​​between adjacent nodes.

[0157] The algorithm determines the nature of the current sudden abnormal event by comparing the current spatial variation coefficient with the baseline value of the historical normal period and combining the significance of the time lag correlation coefficient.

[0158] If the judgment logic output is "local event", characterized by an increased spatial variation coefficient and no significant time lag correlation between nodes, then the gateway will only mark the urgent data of the source node and upload it normally.

[0159] If the judgment logic output is "regional events with spatial diffusion trends", characterized by spatial variation coefficients within a reasonable range but with clear time lag correlations, and outliers appearing sequentially in a specific direction, the gateway will activate the proactive early warning mechanism.

[0160] Under this mechanism, the data aggregation gateway no longer passively waits for instructions from the central processing platform, but immediately generates temporary early warning instruction packages.

[0161] This instruction package includes an event type identifier and suggested monitoring strategy parameters, such as a suggestion to temporarily adjust the sleep cycle to... The recommended duration is 1 hour.

[0162] The gateway broadcasts the temporary warning instruction to all relevant nodes within its communication coverage area, especially those nodes that have not yet shown any abnormalities but are located on the predicted spread path, through its long-range wireless communication module.

[0163] Upon receiving this instruction, the surrounding nodes' event-driven adaptive sleep / wake controllers will temporarily adopt the parameters suggested by the gateway in their subsequent decision functions, overriding the originally calculated parameters. This value allows for early entry into a higher level of monitoring.

[0164] This proactive collaboration at the edge enables forward-looking defense against the spread of potential risks, greatly shortening the overall response time of the system to regional events.

[0165] The central processing platform, deployed on cloud servers or a local high-performance computing center, is the system's intelligent brain. Please refer to the attached document. Figure 1 and attached Figure 5 Logically, the platform includes multiple subsystems such as a data warehouse, a model training engine, an expert knowledge base, a sensor health assessment module, and a visual early warning interface.

[0166] The data warehouse employs a hybrid architecture combining distributed time-series databases and relational databases. All raw monitoring data, node status data, event probability data, and sensor self-diagnostic metadata uploaded from the data aggregation gateway are written to the time-series database in real time and indexed according to node identifiers and timestamps. The relational database stores system configuration information, node metadata, user information, event logs, and more.

[0167] The model training engine is the core of the platform's continuous self-optimization. It is responsible for training and updating the aforementioned lightweight anomaly classifier and the historical normal operating condition parameter model.

[0168] The training process is a continuous closed loop. The platform continuously collects labeled data uploaded by all intelligent monitoring nodes.

[0169] The labeled data not only includes multi-dimensional feature vectors calculated locally at each node, but more importantly, it includes final event labels verified by the platform's expert system or by human review.

[0170] Based on the rules in the expert knowledge base, the expert system combines multi-node and multi-time-scale data correlation analysis to conduct secondary analysis and confirmation of abnormal events reported by nodes, forming high-confidence labels.

[0171] The model training engine retrains existing anomaly classifier models periodically, such as weekly, using newly labeled data accumulated over a period of time.

[0172] Training employs either incremental learning or full retraining strategies.

[0173] After training, the engine uses an independent validation dataset to evaluate the performance of the new model. The evaluation metrics include accuracy, recall, precision, and model size.

[0174] When a new model improves the recall rate compared to the current deployment version, for example, by more than 5%, and the model size does not exceed the node storage limit, the new model is considered to have better performance.

[0175] At this point, the training engine compresses and differentially encodes the parameters of the new model, as well as the historical normal operating condition parameter model updated based on the latest data, and packages them into a software update package. This update package is then wirelessly distributed to each intelligent monitoring node through the data aggregation gateway.

[0176] Under the control of the microcontroller unit, the node receives the update packet during the communication idle period or the next timed wake-up. After backing up the old model, it writes the new model parameters into the specified flash memory area and completes the verification and activation.

[0177] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0178] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-parameter intelligent monitoring system for fishery waters, characterized in that, include: Multiple intelligent monitoring nodes deployed in the target waters, at least one data aggregation gateway, and a central processing platform in the cloud or locally; The intelligent monitoring node includes a sensor array, a microcontroller unit, a wireless communication module, an energy management module, and an event-driven adaptive sleep / wake-up controller. The sensor array includes at least a dissolved oxygen sensor, a pH sensor, a temperature sensor, and an ammonia nitrogen sensor, used to simultaneously collect multi-dimensional water quality parameters; The microcontroller unit has built-in signal conditioning circuitry and an analog-to-digital converter, which are used to filter, amplify and digitize the raw analog signals output by the sensor array. The event-driven adaptive sleep / wake controller is used to execute a sleep decision algorithm based on multi-parameter joint probability inference and dynamic energy budget; The execution process of the hibernation decision algorithm is as follows: the microcontroller unit wakes up and drives the sensor array to complete a full parameter acquisition in a preset basic sampling period, and sends the acquired water quality parameter dataset to the event-driven adaptive hibernation and wake-up controller. The event-driven adaptive sleep / wake controller compares the current dataset with the historical normal operating condition parameter model stored in local memory.

2. The intelligent multi-parameter monitoring system for fishery waters according to claim 1, characterized in that, The event-driven adaptive sleep / wake controller compares the current dataset with the historical normal operating condition parameter model stored in local memory, specifically as follows: Calculate the normalized distance of each parameter value in the current dataset from the center value of its historical normal working condition probability distribution interval, and calculate the comprehensive deviation index based on the preset weight coefficients. Simultaneously, the matching degree between the relationship between each parameter within the current parameter set and the parameter coupling relationship defined in the historical model is calculated to obtain the relationship consistency index; The comprehensive deviation index and the relationship consistency index together constitute a multi-dimensional feature vector; The event-driven adaptive sleep / wake controller inputs the multidimensional feature vector into a localized lightweight anomaly classifier; The localized lightweight anomaly classifier is a machine learning model pre-trained and distributed by the central processing platform. Its output is the probability of the occurrence of a sudden anomaly event in the water quality state at the current sampling time. The energy management module monitors the remaining power of the node battery in real time and calculates a dynamic energy budget coefficient that is positively correlated with the remaining power. The event-driven adaptive sleep / wake controller ultimately calculates the duration of the next sleep cycle based on the event occurrence probability and the dynamic energy budget coefficient through a determined decision function. The logic of the decision function is: the higher the probability of an event occurring, the shorter the calculated duration of the next sleep cycle. The lower the dynamic energy budget coefficient, the longer the calculated duration of the next hibernation cycle, but there is an upper limit set by the probability of the event occurring.

3. The intelligent multi-parameter monitoring system for fishery waters according to claim 2, characterized in that, The specific execution rules of the decision function are as follows: Set high probability thresholds and low probability thresholds; when the probability of an event occurring is greater than or equal to the high probability threshold, regardless of the value of the dynamic energy budget coefficient, force the next sleep cycle to be set to an extremely short active monitoring cycle. When the probability of an event occurring is less than or equal to the low probability threshold, the sleep duration is adjusted within the preset normal range mainly based on the dynamic energy budget coefficient. When the probability of an event occurring is between a low probability threshold and a high probability threshold, a linear interpolation function is used to smoothly transition the sleep duration from the value corresponding to the low probability threshold to the active monitoring cycle corresponding to the high probability threshold.

4. The intelligent multi-parameter monitoring system for fishery waters according to claim 3, characterized in that, The training and update mechanism of the localized lightweight anomaly classifier is as follows: The central processing platform continuously collects labeled data uploaded by all intelligent monitoring nodes. The labeled data includes multi-dimensional feature vectors and final event labels confirmed by the platform's expert system or by manual review. The central processing platform periodically retrains the anomaly classifier model using accumulated labeled data; When the performance evaluation metrics of the new model are better than those of the old model currently deployed on the node, the central processing platform packages the new model parameters and the corresponding historical normal operating condition parameter model, and distributes them to each intelligent monitoring node through the data aggregation gateway in a differential update manner. After receiving the update package, the intelligent monitoring node, under the control of the microcontroller unit, completes the replacement and verification of model parameters during idle periods.

5. The intelligent multi-parameter monitoring system for fishery waters according to claim 4, characterized in that, The data aggregation gateway has edge collaborative computing capabilities; When a smart monitoring node enters a high-frequency sampling mode due to a persistently high probability of an event, the node sends a collaborative computing request flag along with the data packet it uploads to the gateway. Upon receiving the flag, the data aggregation gateway actively retrieves recent historical data from neighboring nodes around the requesting node. The data aggregation gateway has a built-in edge computing module, which runs a spatial correlation analysis algorithm to determine whether a sudden abnormal event is a local event or a regional event with a spatial diffusion trend. If the analysis results indicate that the anomaly has spatial diffusion, the data aggregation gateway directly broadcasts a temporary warning instruction containing a suggested shortened sleep cycle parameter to all relevant nodes within its communication range. Upon receiving a temporary warning command, the surrounding nodes' event-driven adaptive sleep / wake-up controllers temporarily adopt the parameters suggested by the gateway in their decision functions.

6. The intelligent multi-parameter monitoring system for fishery waters according to claim 5, characterized in that, The central processing platform includes a data warehouse, a model training engine, an expert knowledge base, and a visual early warning interface; The central processing platform performs cross-node trend risk assessment, the process of which is as follows: Periodically analyze the data of all nodes to identify parameter change patterns that are spatially clustered and temporally trending, and match these patterns with event patterns in the expert knowledge base; When the matching degree exceeds a set threshold, the central processing platform determines that there is a systemic risk and issues a global monitoring strategy enhancement instruction to all nodes in the affected area through the data aggregation gateway, and uniformly adjusts the decision function parameters of the relevant nodes.

7. The intelligent multi-parameter monitoring system for fishery waters according to claim 6, characterized in that, The sensor array of the intelligent monitoring node adopts a modular packaging and interface design; The microcontroller unit is connected to each sensor module through a standardized digital interface; each sensor module integrates a self-diagnostic circuit for periodically detecting the sensitivity, zero-point drift, and response time of the sensor probe. The self-diagnostic results are uploaded as metadata attached to the regular water quality data package; The central processing platform is equipped with a sensor health assessment module, which analyzes the rationality of long-term uploaded self-diagnostic data and water quality parameter data to determine whether the sensor has experienced performance degradation or malfunction.

8. The intelligent multi-parameter monitoring system for fishery waters according to claim 1, characterized in that, The system's communication protocol employs a mechanism combining time division multiple access and frequency hopping. The data aggregation gateway is responsible for allocating communication time slots to all intelligent monitoring nodes within its jurisdiction. When the event-driven adaptive sleep-wake controller decides to wake up and send data, it must first synchronize its time slots with the gateway. When the channel is interfered with, the wireless communication module automatically switches to a preset backup frequency point for communication according to the protocol.

9. The intelligent multi-parameter monitoring system for fishery waters according to claim 8, characterized in that, For urgent data packets marked as having a high probability of occurrence, the system assigns them the highest communication priority, allowing them to access the channel in a contention-based manner in unspecified time slots and enjoy redundant transmission guarantees.

10. A multi-parameter intelligent monitoring system for fishery waters according to claim 9, characterized in that, The specific method by which the energy management module calculates the dynamic energy budget coefficient is as follows: The ratio of the current remaining battery power to the total battery capacity is obtained by mapping it through a preset smoothing function.

Images