Aquaculture multi-device cooperative control system based on industrial internet of things

Abstract

The application discloses an aquaculture multi-device cooperative control system based on an industrial internet of things, and belongs to the technical field of the industrial internet of things. The system comprises a data acquisition module, which is used for responding to heterogeneous device data streams converged by an industrial internet of things gateway, and constructing a multi-modal time series dataset containing dissolved oxygen, temperature and device state parameters; a time domain alignment module, which generates a synchronous data frame by performing sliding window time domain alignment on the multi-modal time series dataset; a confidence inference module, which quantifies the uncertainty of the synchronous data frame by using a confidence rule base based on evidence reasoning, and obtains a parameterized feature vector; a game decision module, which inputs the parameterized feature vector into a multi-device cooperative decision model based on cooperative game theory, and calculates an optimal control instruction set that makes the comprehensive energy consumption and water quality regulation error reach a Nash equilibrium; and an instruction distribution module, which issues time stamp synchronized differentiated execution instructions to an oxygenator, a bait feeder and a water pump through the industrial internet of things gateway.

Description

Technical Field

[0001] This invention relates to the field of industrial Internet of Things (IoT) technology, specifically to a multi-equipment collaborative control system for aquaculture based on industrial IoT. Background Technology

[0002] Water quality parameter regulation and collaborative control of multiple types of equipment in aquaculture environments face multiple challenges, including data heterogeneity, inconsistent time scales, and decision-making uncertainty. Existing solutions generally employ single-parameter threshold triggering or rule-based sequential control. These methods collect environmental and equipment status information separately using dissolved oxygen sensors, temperature sensors, and current transformers, and then perform simple data fusion before using preset logic tables or PID controllers to drive actuators such as aerators, feeders, and water pumps. This approach relies heavily on deterministic models or hard threshold judgments, failing to effectively address the uncertainties arising from the dynamic coupling of controlled objects, sensor noise, and equipment execution delays in aquaculture.

[0003] The main shortcomings of existing technical solutions lie in two aspects. Firstly, they lack a mechanism for quantifying and characterizing the inherent uncertainties in monitoring data and equipment status information. Sensors are affected by factors such as water turbidity, biological attachment, and electromagnetic interference, resulting in values ​​often accompanied by random biases and occasional anomalies. Parameter drift caused by changes in the operating point of the equipment itself also introduces non-negligible state ambiguity. Traditional methods directly feed these observations as deterministic inputs into the decision-making process, leading to false triggering or control saturation near boundaries in control logic based on clear thresholds and rules, increasing the risk of water quality fluctuations and ineffective energy consumption. Secondly, multi-device control decisions typically optimize aeration, feeding, and water circulation functions separately as independent loops, or employ centralized single-objective optimization, failing to consider the interactive constraints between energy consumption and control effectiveness from a game theory perspective. This approach oversimplifies the nonlinear impact of the joint strategy formed by the actuators working collaboratively on the overall optimization objective, making it difficult to achieve a balance between reducing the overall energy consumption of multiple devices and maintaining precise control of water quality parameters.

[0004] The aforementioned problems faced in aquaculture stem from the lack of a collaborative control architecture capable of consistently and formally representing the uncertainties in multimodal time-series data and seeking equilibrium strategies among competing optimization objectives. How to handle the transmission of sensor information uncertainty to multi-device joint decision-making, and how to achieve a trade-off between individual device rationality and overall energy conservation and emission reduction, have become pressing technical challenges that need to be overcome. Summary of the Invention

[0005] This invention provides a multi-device collaborative control system for aquaculture based on the Industrial Internet of Things, aiming to solve the problems of lack of quantification of uncertainty in multimodal monitoring data and the inability of multi-device collaborative decision-making to simultaneously achieve optimal comprehensive energy consumption and water quality control accuracy in existing technologies.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a multi-device collaborative control system for aquaculture based on the Industrial Internet of Things. By combining confidence reasoning with game theory decision-making, it solves the problems of delayed response, poor coordination, and difficulty in handling data uncertainty in traditional aquaculture multi-device control, thereby reducing overall energy consumption and improving the accuracy of water quality control.

[0007] The system comprises a data acquisition module, a time-domain alignment module, a confidence inference module, a game-theoretic decision-making module, and an instruction distribution module. The data acquisition module responds to heterogeneous device data streams aggregated by the industrial IoT gateway, constructing a multimodal time-series dataset containing dissolved oxygen, temperature, and device status parameters. This ensures that data from devices with different sources and protocols can be uniformly captured and represented. The time-domain alignment module performs a sliding window-based time-domain alignment operation on the multimodal time-series dataset, generating synchronized data frames with a unified timestamp reference. This eliminates time inconsistencies caused by differences in sampling frequencies and transmission delays between devices, providing spatiotemporally consistent input for subsequent inference and decision-making. The confidence inference module uses a confidence rule base based on evidence-based reasoning to quantify the uncertainty of each data point in the synchronized data frame, obtaining a parameterized feature vector with a confidence distribution. This explicitly expresses sensor measurement errors and device status ambiguities, enabling the system to robustly cope with data drift and noise interference in real-world environments. The game theory decision-making module inputs parameterized feature vectors into a multi-device collaborative decision-making model based on cooperative game theory. It calculates the optimal control command set that achieves Nash equilibrium between the overall energy consumption and water quality control error of multiple aquaculture devices. This allows aerators, feeders, and pumps to spontaneously coordinate power allocation based on their respective revenue drivers, preventing overall imbalance caused by over-optimization of a single indicator. The command distribution module, based on this optimal control command set, issues differentiated execution commands, synchronized with timestamps, to the aerators, feeders, and pumps via an industrial IoT gateway. This ensures that multiple devices execute synchronously under a unified time reference, thereby eliminating the weakening effect of water quality regulation caused by misaligned action sequences.

[0008] As a technical solution of this invention, when constructing a multimodal time-series dataset, the data acquisition module subscribes to the current signal of the aerator, the rotation speed signal of the feeder, and the vibration signal of the water pump through the data aggregation layer of the industrial IoT gateway, and performs protocol-independent raw data capture operations on each signal. The analog signals in the captured heterogeneous device data streams are converted into digital signals using an analog-to-digital converter, and a timestamp is added to the digital signals. According to the dissolved oxygen sensor, temperature sensor, and the data source identifier corresponding to each device, the timestamp-tagged digital signals are categorized and stored in a circular buffer of the in-memory database, forming a multimodal time-series dataset. This processing method avoids timing jitter caused by protocol conversion and format parsing, ensuring high fidelity of the time-series data.

[0009] Preferably, the temporal alignment module reads a multimodal time-series dataset of a preset duration from the circular buffer of the in-memory database, identifies the starting data point with the earliest timestamp label, and slides backward with a fixed window step size, extracting all data points intersecting with the window's start and end times at each sliding window position. Linear interpolation is performed on all data points with the same timestamp label extracted within each sliding window to ensure a consistent number of data points output by each sliding window, and all data points associated with that sliding window are packaged into a synchronization data frame. Thus, sensor and device data with different sampling frequencies are regularized into synchronization frames of equal duration and density, providing structured input for subsequent confidence inference.

[0010] During uncertainty quantification, the confidence inference module matches dissolved oxygen, temperature, and equipment status parameters from the synchronized data frame with multiple preset reference levels in the confidence rule base, calculating an initial confidence assignment value for each reference level. All calculated initial confidence assignment values ​​are then vectorized and concatenated according to their respective synchronized data frames to generate a parameterized feature vector where each dimension corresponds to a reference level and the sum of confidence values ​​across dimensions is one. This confidence distribution representation ensures that the system's incomplete uncertainty regarding any data point is preserved and transmitted to the decision-making stage, avoiding information loss caused by hard thresholding.

[0011] Furthermore, when the dissolved oxygen value falls within the overlapping threshold range of two adjacent reference levels, the confidence assignment values ​​for the low and medium dissolved oxygen levels are calculated based on the ratio of the distances from the dissolved oxygen value to the center values ​​of the two reference levels, while the confidence assignment value for the high dissolved oxygen level is set to zero. In this way, small fluctuations in sensor readings near the level boundaries will not cause abrupt changes in assignment, enhancing the stability of the control decision.

[0012] The game-theoretic decision-making module defines aerators, feeders, and water pumps as different players in a cooperative game model, and uses the confidence distribution in the parameterized feature vector of each player as their private information type. A strategy space containing multiple discrete power output levels is defined for each player, along with a payoff function. This payoff function calculates the difference between water quality improvement benefits and energy consumption costs based on the combined strategies adopted by all players. An iterative optimal response algorithm is executed, allowing each player to choose the strategy that maximizes its own payoff function given the current strategies of other players, until all players no longer change their strategies. The resulting strategy combination is taken as the Nash equilibrium solution. The discrete power output level corresponding to each player in the Nash equilibrium solution is converted into control commands containing device identifiers, target power values, and durations, forming the optimal control command set. This mechanism allows each device to naturally achieve a balance between global energy consumption and water quality control errors while pursuing its own maximum benefit, and the algorithm converges to Nash equilibrium, ensuring the stability and repeatability of the control commands.

[0013] In the specific iterative solution, the initial power levels of the aerator, feeder, and water pump are randomly initialized as the current strategy combination. In the first iteration, the strategies of the feeder and water pump are fixed, all possible power levels of the aerator are enumerated, and the power level that maximizes the aerator's benefit is selected as the new strategy. In the second iteration, the new strategy of the aerator and the original strategy of the water pump are fixed, and the strategy of the feeder is optimized. In the third iteration, the new strategies of the aerator and feeder are fixed, and the strategy of the water pump is optimized. The above multiple iterations are repeated until the power levels of all participants no longer change in two consecutive rounds, and the strategy combination obtained in the last iteration is taken as the Nash equilibrium solution. This method has a fast convergence speed and is suitable for the limited computing resources of industrial gateways.

[0014] The command distribution module extracts control commands for the aerator, feeder, and water pump from the optimal control command set, encapsulates them into message frames conforming to their respective communication protocols, appends the same global execution timestamp to each message frame, and pushes them to the command sending queue of the industrial IoT gateway. Through multiple parallel output channels of the industrial IoT gateway, each message frame is simultaneously sent at the time indicated by the global execution timestamp. Preferably, the global execution timestamp uses network time synchronized based on the IEEE 1588 precise time protocol, ensuring that the time synchronization error between multiple parallel output channels is less than 1 millisecond, thereby guaranteeing strict synchronization of multiple device actions at the physical execution level.

[0015] In a more specific implementation, the real-time operating system scheduler of the industrial IoT gateway creates a synchronous transmission task, reads a message frame with a globally executed timestamp appended from the command transmission queue, and suspends the task. The scheduler periodically checks whether the current system clock has reached the time indicated by the globally executed timestamp. Once it has, it simultaneously wakes up the first coroutine associated with the aerator output channel, the second coroutine associated with the feeder output channel, and the third coroutine associated with the water pump output channel. Each coroutine drives its corresponding output channel to complete the message frame transmission. This method, utilizing coroutines for concurrent transmission, effectively reduces timing deviations.

[0016] In a preferred embodiment of the present invention, the system further includes a feedback update module. After the equipment executes the control command, the industrial IoT gateway receives execution completion confirmation frames from the aerator, feeder, and water pump, and extracts the actual execution time and actual execution power value from them; it compares the actual execution time with the expected execution time to obtain the execution time deviation, and compares the actual execution power value with the target power value to obtain the execution power deviation; the above deviations are input as feedback signals to the parameter updater of the multi-device collaborative decision-making model, and the parameter updater adjusts the energy consumption cost coefficient in the payoff function of each game participant accordingly, and uses the adjusted coefficients for the calculation of the next control command set. Through this feedback loop, the system can adapt to equipment aging, operating condition drift, and environmental changes, maintaining the optimality of collaborative control.

[0017] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0018] By constructing a confidence rule base based on evidence-based reasoning, the uncertainty of synchronous data frames composed of dissolved oxygen, temperature, and equipment status parameters is systematically quantified. Each measurement value in the synchronous data frame is matched with multiple preset reference levels in the confidence rule base. Based on the relative distance between the measured value and the threshold interval or the overlapping part of the interval, an initial confidence level is assigned to each reference level. All confidence level assignments are then concatenated into a parameterized feature vector where each dimension corresponds to a reference level and the sum of confidence levels between dimensions is one. This approach does not rigidly classify the original sensor readings but explicitly preserves the fuzzy attribute of measured values ​​falling on the boundaries of different reference intervals in the form of confidence level distribution. This allows subsequent decision-making processes to perceive the credibility differences of noise, drift, and random anomalies in the data. By directly using the confidence level distribution vector instead of scalar threshold triggering, frequent switching or malfunctions of control modes due to small fluctuations are avoided when parameters such as dissolved oxygen and temperature are in the transition range of the reference level. This suppresses overshoot and oscillations in the transient process of water quality control, thereby reducing the additional energy consumption caused by ineffective start-up and shutdown of equipment while maintaining the stability of the water environment.

[0019] By introducing a multi-device collaborative decision-making model based on cooperative game theory, aerators, feeders, and water pumps are treated as game participants. The confidence distribution in the parameterized feature vectors of each device is used as the private information type of the participants. A strategy space containing discretized power output levels and a payoff function that combines water quality improvement benefits with energy consumption costs are defined for each participant. Then, an iterative optimal response algorithm is used to search for a Nash equilibrium solution. This mechanism ensures that no single device has an incentive to unilaterally change its own power level when the strategies of other devices are fixed. The resulting optimal control instruction set is at a stable equilibrium point in the multi-device joint strategy space. This approach avoids the limitations of centralized optimization, which requires constructing an explicit global objective function and relying on complex weight tuning. By converging to a Nash equilibrium within the cooperative game framework, the combined energy consumption of multiple devices and water quality control errors are simultaneously considered. This allows the final strategy to reduce the total power consumption of aeration, feeding, and water circulation while maintaining the joint deviation of dissolved oxygen and temperature within an acceptable range, achieving an autonomous balance between energy saving and control accuracy under coordinated operation of multiple devices. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0021] Figure 1 This is a schematic diagram of a multi-equipment collaborative control system for aquaculture based on the Industrial Internet of Things.

[0022] Figure 2 This is a flowchart of the time-domain alignment and synchronous framing of multimodal industrial sensor data;

[0023] Figure 3 This is a flowchart of the parameterized feature vector generation process for the confidence inference module;

[0024] Figure 4 This is a schematic diagram of multimodal time series data acquisition;

[0025] Figure 5 It is a synchronous interpolation curve of multimodal time series data;

[0026] Figure 6 It is the energy consumption cost coefficient update curve of the multi-device collaborative decision-making model. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] See Figure 1 This invention provides a collaborative control system for multiple aquaculture devices based on the Industrial Internet of Things (IIoT). The specific implementation of this system includes a data acquisition module, a time-domain alignment module, a confidence inference module, a game theory decision-making module, and an instruction distribution module. The data acquisition module responds to the heterogeneous device data streams aggregated by the IIoT gateway, constructing a multimodal time-series dataset containing dissolved oxygen, temperature, and device status parameters. The time-domain alignment module performs a sliding window-based time-domain alignment operation on the multimodal time-series dataset, generating synchronized data frames with a unified timestamp reference. The confidence inference module uses a confidence rule base based on evidence-based reasoning to perform uncertainty quantification on each data point in the synchronized data frame, obtaining a parameterized feature vector with a confidence distribution. The game theory decision-making module inputs the parameterized feature vector into a multi-device collaborative decision-making model based on cooperative game theory, calculating the optimal control instruction set that achieves Nash equilibrium between the comprehensive energy consumption and water quality control error of multiple aquaculture devices. Based on the optimal control instruction set, the instruction distribution module issues differentiated execution instructions synchronized with timestamps to the aerator, feeder, and water pump through the IIoT gateway.

[0029] Example 1:

[0030] In specific implementation, please refer to Figure 2 The data aggregation layer of the industrial IoT gateway subscribes to the current signals of the aerator, the speed signals of the feeder, and the vibration signals of the water pump, and performs protocol-independent raw data acquisition operations on these signals. This protocol-independent raw data acquisition operation is implemented through the multi-protocol adaptation engine built into the industrial IoT gateway, which supports parallel parsing of four industrial communication protocols: ModbusTCP, PROFINET, EtherCAT, and OPCUA. During the subscription process, the data aggregation layer of the industrial IoT gateway sends corresponding protocol subscription request frames to the communication interfaces of the aerator, feeder, and water pump. After receiving the subscription confirmation frame, it continuously receives the raw data streams periodically reported by the aerator, feeder, and water pump, without performing any protocol-layer trimming on the raw data streams, fully preserving the physical layer payload.

[0031] The captured heterogeneous device data streams contain both analog and digital signals. The analog signals are converted to digital signals using an analog-to-digital converter (ADC). The ADC is a 16-bit successive approximation type, with a sampling frequency set to 10kHz to cover harmonic components within 200Hz in the aerator current signal, the pulse-width modulation fundamental frequency within 500Hz in the feeder speed signal, and the bearing characteristic frequency within 1kHz in the water pump vibration signal. After the ADC conversion, a timestamp is added to all digital signals. The timestamp is generated based on the IEEE 1588 precision time protocol clock locally maintained by the industrial IoT gateway, with timestamps accurate to the microsecond level and including both second and nanosecond fields. The timestamped digital signals are categorized and stored in a circular buffer of the in-memory database according to five data source identifiers: dissolved oxygen sensor, temperature sensor, aerator device identifier, feeder device identifier, and water pump device identifier. The circular buffer consists of five pre-allocated circular queues within the in-memory database. Each circular queue has a capacity of 2048 data units. Each data unit stores a single sampled digital signal and its timestamp label. When any circular queue is full, the newly arrived data unit overwrites the oldest data unit. The data stored in the five circular queues, sorted by timestamp label, together constitute the multimodal time series dataset.

[0032] The time-domain alignment module reads a multimodal time-series dataset of a preset duration (5 seconds) from the circular buffer of the in-memory database. During reading, a batch reading method is used, retrieving all data units generated within the past 5 seconds from the circular queues of the dissolved oxygen sensor, temperature sensor, aerator, feeder, and water pump. It then identifies the starting data point in the multimodal time-series dataset with the earliest timestamp label; this starting data point is the data unit with the smallest timestamp label among all five circular queues.

[0033] Using the starting data point as a reference, the window slides backward with a fixed step size of 100 milliseconds. At each sliding window position, all data points intersecting with the window's start and end times are extracted. The width of the sliding window is set to 500 milliseconds. For a window with a start time of... And the end time is A sliding window of milliseconds iterates through the timestamp labels in each circular queue. and All data units between them are incorporated into the original data set of the current sliding window.

[0034] Linear interpolation is performed on all data points under the same timestamp label extracted within each sliding window to ensure that the number of data points output by each sliding window is consistent. The linear interpolation process is as follows: the 500-millisecond interval within the sliding window is divided into 51 target time points at 10-millisecond intervals. For each target time point... Find the nearest timestamp label in the original dataset of each data source. The previous data point and the next data point Using linear interpolation formula

[0035]

[0036] Calculate the target time point Corresponding interpolation results . In the formula, This represents the value of the previous data point. This indicates the value of the next data point. This represents the timestamp label of the previous data point. Indicates the timestamp label of the next data point. This indicates the target time point to be interpolated. The value range is in 10-millisecond increments from the start time of the window. Start by gradually increasing until the window ends. The linear interpolation operation described above is performed on all five data sources. Each sliding window ultimately generates 51 data points at a unified time point. Each time point corresponds to a vector containing five values: dissolved oxygen interpolation, temperature interpolation, aerator status interpolation, feeder status interpolation, and water pump status interpolation. All data points associated with this sliding window are packaged into a synchronization data frame. The synchronization data frame is a 51-row, 5-column matrix, where each row corresponds to a target time point and each column corresponds to a data source. The time-domain alignment module continuously slides the window, generating a synchronization data frame for each window, forming a sequence of synchronization data frames for the confidence inference module to read.

[0037] See Figure 4 The horizontal axis of the graph represents time in seconds, ranging from 0 to 5 seconds. The left side of the vertical axis shows dissolved oxygen concentration (mg / L), temperature (degrees Celsius), and the effective value of the aerator current signal (amperes). The right side of the vertical axis shows the feeder speed deviation rate (percentage) and water pump vibration intensity (mm / s). The legend identifies five curves, corresponding to dissolved oxygen concentration, temperature, effective value of aerator current signal, feeder speed deviation rate, and water pump vibration intensity, respectively.

[0038] The dissolved oxygen concentration curve showed a stable fluctuation trend, with the overall range remaining between 5.4 and 6.0 mg / L, indicating that the dissolved oxygen in the water body remained within the threshold range of the medium dissolved oxygen level, which meets the definition of the dissolved oxygen reference level in Example 2. The temperature curve was also relatively stable, with minimal fluctuations, consistently remaining within the threshold range of the high temperature level of approximately 25 degrees Celsius, meeting the high temperature level standard of the temperature reference level.

[0039] The effective value curve of the aerator current signal remained between approximately 19 and 21 amperes, exhibiting good overall stability, indicating that the aerator was operating within the normal range (according to the threshold definition of the aerator's normal operating state level in Example 2). The feeder speed deviation rate curve fluctuated significantly, ranging from approximately 0% to 25%, mostly below 15%, with occasional short-term peaks exceeding 15%, reflecting that the feeder was in an abnormal operating state at certain times, consistent with the classification of normal and abnormal levels of feeder speed deviation rate in Example 2.

[0040] The pump vibration intensity curves show significant fluctuations, ranging from approximately 1.5 to 3.5 mm / s, all below the abnormal state threshold of 7.1 mm / s, indicating that the pump vibration intensity remains within the normal state range. Overall, all five curves exhibit stable and continuous data changes within 5 seconds, reflecting that the multimodal time-series dataset captured by the industrial IoT gateway in Example 1, after being processed by the time-domain alignment module, achieves synchronization and effective acquisition of multi-device data at the microsecond-level timestamp.

[0041] The data shown in the figure is multimodal time-series data sorted by timestamp in the circular buffer used in Example 1. After processing by the time-domain alignment module based on sliding window and linear interpolation methods, it forms a synchronized data frame, providing an accurate and synchronized raw data foundation for the uncertainty quantification and parameterized feature vector generation of the confidence inference module in Example 2. The continuous fluctuation trend of dissolved oxygen concentration and temperature data in the figure, as well as the dynamic changes of equipment status signals, demonstrate the system's high-precision real-time acquisition capability of signals collected from multiple devices, ensuring the timeliness and accuracy of data input for the multi-device collaborative control system.

[0042] Example 2:

[0043] In specific implementation, please refer to Figure 3The confidence inference module obtains a synchronization data frame from the time-domain alignment module. The synchronization data frame is a 51-row, 5-column numerical matrix, with each row corresponding to a target time point and each column containing, in order, the dissolved oxygen interpolation, temperature interpolation, aerator status interpolation, feeder status interpolation, and feeder speed status interpolation (i.e., water pump status interpolation). The confidence inference module extracts each set of data points from the synchronization data frame row by row and performs uncertainty quantization processing on each data point individually to generate the parameterized feature vector corresponding to that data point.

[0044] The confidence rule base is pre-built and stored in the non-volatile storage area of ​​the confidence inference module. The confidence rule base consists of three parts: dissolved oxygen reference level, temperature reference level, and equipment status reference level. The dissolved oxygen reference level is divided into three levels: low dissolved oxygen, medium dissolved oxygen, and high dissolved oxygen. The threshold range for the low dissolved oxygen level is a dissolved oxygen concentration greater than or equal to 0 mg / L and less than 4.0 mg / L, with a center value of 2.0 mg / L. The threshold range for the medium dissolved oxygen level is a dissolved oxygen concentration greater than 3.5 mg / L and less than or equal to 6.5 mg / L, with a center value of 5.0 mg / L. The threshold range for the high dissolved oxygen level is a dissolved oxygen concentration greater than 6.0 mg / L and less than or equal to 10.0 mg / L, with a center value of 8.0 mg / L. The temperature reference levels are divided into three categories: low temperature, medium temperature, and high temperature. The threshold range for the low temperature level is 0°C or higher and less than 18°C, with a center value of 12°C. The threshold range for the medium temperature level is 16°C or higher and less than or equal to 28°C, with a center value of 22°C. The threshold range for the high temperature level is 26°C or higher and less than or equal to 45°C, with a center value of 32°C. Equipment status reference levels are set separately for aerators, feeders, and water pumps. Each piece of equipment corresponds to two reference levels: a normal status level and an abnormal status level. The threshold range for the normal state level of an aerator is that the effective value of the aerator current signal is greater than or equal to 0 Amperes and less than 1.2 times the rated current of the aerator; the threshold range for the abnormal state level of an aerator is that the effective value of the aerator current signal is greater than or equal to 1.2 times the rated current of the aerator. The threshold range for the normal state level of a feeder is that the absolute value of the feeder speed deviation rate is less than 15%; the threshold range for the abnormal state level of a feeder is that the absolute value of the feeder speed deviation rate is greater than or equal to 15%. The threshold range for the normal state level of a water pump is that the water pump vibration intensity value is less than 7.1 mm / s; the threshold range for the abnormal state level of a water pump is that the water pump vibration intensity value is greater than or equal to 7.1 mm / s.

[0045] For a dissolved oxygen value within a data point in a synchronized data frame, the confidence inference module compares the dissolved oxygen value with threshold ranges for low, medium, and high dissolved oxygen levels. When the dissolved oxygen value falls within the non-overlapping portion of the low and medium dissolved oxygen level threshold ranges, the confidence inference module sets the confidence assignment value for the dissolved oxygen reference level it falls into to 1, and sets the confidence assignment values ​​for the other two dissolved oxygen reference levels to 0. When the dissolved oxygen value falls within the overlapping region of the low and medium dissolved oxygen level threshold ranges (i.e., the dissolved oxygen value is greater than 3.5 mg / L and less than 4.0 mg / L), or when the dissolved oxygen value falls within the overlapping region of the medium and high dissolved oxygen level threshold ranges (i.e., the dissolved oxygen value is greater than 6.0 mg / L and less than 6.5 mg / L), the confidence inference module assigns confidence to the two adjacent dissolved oxygen reference levels based on the ratio of the distances from the dissolved oxygen value to the center values ​​of the two adjacent dissolved oxygen reference levels, and sets the confidence assignment value for the third dissolved oxygen reference level to 0. Taking a dissolved oxygen value located in the overlapping region of low and medium dissolved oxygen levels as an example, the confidence inference module calculates the absolute value of the distance from the dissolved oxygen value to the center value of the low dissolved oxygen level. and the absolute value of the distance from the dissolved oxygen value to the center value of the medium dissolved oxygen level. ,in , Then, calculate the confidence allocation value for the low dissolved oxygen level according to the following formula. :

[0046]

[0047] In the formula, This represents the absolute value of the distance from the dissolved oxygen level to the center value of 2.0 mg / L at the lowest dissolved oxygen level. This represents the absolute value of the dissolved oxygen level from the median dissolved oxygen level of 5.0 mg / L. This represents the confidence level assignment for low dissolved oxygen levels. The confidence level assignment for medium dissolved oxygen levels is also shown. Depend on The confidence assignment values ​​for high dissolved oxygen levels were obtained. Set to 0. When the dissolved oxygen value is in the overlapping area of ​​the medium and high dissolved oxygen levels, the confidence inference module calculates the confidence assignment values ​​for the medium and high dissolved oxygen levels in the same way, and sets the confidence assignment value for the low dissolved oxygen level to 0.

[0048] For temperature values ​​within the same data point, the confidence inference module performs a similar processing procedure to that for dissolved oxygen values. The module compares the temperature value with the threshold ranges for low, medium, and high temperature levels. When the temperature value falls entirely within the non-overlapping threshold range of a temperature reference level, the confidence score for that reference level is set to 1, and the confidence scores for the other two reference levels are set to 0. When the temperature value is located in the overlapping area of ​​two adjacent temperature reference levels, the confidence score is assigned based on the ratio of the distances from the temperature value to the center values ​​of the two adjacent reference levels, and the confidence score for the third reference level is set to 0. The distances from the temperature value to the center values ​​of the low, medium, and high temperature levels are calculated using the absolute values ​​of the differences between the temperature value and 12 degrees Celsius, 22 degrees Celsius, and 32 degrees Celsius, respectively.

[0049] For equipment status parameters within the same data point, the confidence inference module processes the effective value of the aerator current signal, the feeder speed deviation rate, and the water pump vibration intensity value separately. The effective value of the aerator current signal is compared with the threshold ranges for the aerator's normal and abnormal status levels. When the effective value of the current signal is less than 1.2 times the aerator's rated current, the confidence allocation value for the aerator's normal status level is set to 1, and the confidence allocation value for the abnormal status level is set to 0; otherwise, the confidence allocation value for the aerator's normal status level is set to 0, and the confidence allocation value for the abnormal status level is set to 1. Similarly, the feeder speed deviation rate is compared with the threshold ranges for the feeder's normal and abnormal status levels. When the absolute value of the speed deviation rate is less than 15%, the confidence allocation value for the feeder's normal status level is set to 1, and the confidence allocation value for the abnormal status level is set to 0; otherwise, the confidence allocation value for the feeder's normal status level is set to 0, and the confidence allocation value for the abnormal status level is set to 1. The vibration intensity value of the water pump is compared with the threshold range of the normal state level and the abnormal state level of the water pump. When the vibration intensity value is less than 7.1 mm / s, the confidence assignment value of the normal state level of the water pump is set to 1, and the confidence assignment value of the abnormal state level is set to 0; otherwise, the confidence assignment value of the normal state level of the water pump is set to 0, and the confidence assignment value of the abnormal state level is set to 1.

[0050] The confidence inference module vectorizes all the initial confidence assignments calculated above according to the same data point, resulting in a temporary vector. This temporary vector contains 12 dimensions: low dissolved oxygen level confidence assignment, medium dissolved oxygen level confidence assignment, high dissolved oxygen level confidence assignment, low temperature level confidence assignment, medium temperature level confidence assignment, high temperature level confidence assignment, aerator normal state level confidence assignment, aerator abnormal state level confidence assignment, feeder normal state level confidence assignment, feeder abnormal state level confidence assignment, water pump normal state level confidence assignment, and water pump abnormal state level confidence assignment. The confidence inference module then performs L1 norm normalization on the temporary vector, specifically calculating the sum of all 12 dimension values. Divide the value of each dimension in the temporary vector by A 12-dimensional parameterized feature vector is generated, where all dimensions of the parameterized feature vector have values ​​greater than or equal to 0 and their sum is equal to 1. The confidence inference module performs the above processing on 51 data points in the synchronous data frame sequentially, generating 51 parameterized feature vectors, forming a parameterized feature vector sequence, which is then output to the game decision module.

[0051] See Figure 5 In the figure, the left side of the vertical axis corresponds to the interpolated values ​​of dissolved oxygen (mg / L), temperature (°C), and feeder status (%), respectively. The right side of the vertical axis corresponds to the interpolated values ​​of aerator status (Amperes) and water pump status (mm / s), respectively. The horizontal axis represents the target time point sequence number, reflecting the sequence number of the 51 target time points in the synchronization data frame output by the time-domain alignment module. The five curves in the figure represent the temporal variation characteristics of dissolved oxygen, temperature, aerator current, feeder speed deviation rate, and water pump vibration intensity after linear interpolation of the multimodal time-series dataset collected by the industrial IoT gateway based on the time-domain alignment module.

[0052] As shown in the figure, the dissolved oxygen interpolation curve exhibits a trend of first rising and then falling, with a peak value of approximately 7.5 mg / L, falling within the threshold range of medium and high dissolved oxygen levels defined in Patent Example 2. The temperature interpolation curve decreases slowly overall, from approximately 25 degrees Celsius to approximately 22 degrees Celsius, falling within the threshold range of the medium temperature level, which conforms to the temperature reference level definition range in Example 2. The aerator status interpolation (current signal) fluctuates significantly, remaining between 0.5 amps and 3.5 amps, with the value not exceeding 1.2 times the aerator's rated current, indicating a normal operating condition. The feeder status interpolation (speed deviation rate) shows an overall fluctuating trend of first decreasing and then increasing, with the lowest point close to 0% and the highest close to 25%. The range below 15% corresponds to a normal operating condition, while the range above 15% corresponds to an abnormal operating condition, indicating that the feeder may experience abnormal operation during certain periods. The pump status interpolation (vibration intensity) fluctuates periodically, with a peak value of approximately 3 mm / s, all of which are below the threshold of 7.1 mm / s for the abnormal state level, indicating that it is in the normal state level.

[0053] Overall, the five multimodal data sets are synchronized and packaged into a 51-row, 5-column synchronized data frame. This demonstrates that the time-domain alignment module achieves time synchronization processing of different time-series signals through sliding windows and linear interpolation methods, generating multimodal data vectors under a unified timestamp benchmark. This data provides an accurate and continuous input basis for the uncertainty quantification of the subsequent confidence inference module, meeting the real-time monitoring requirements for multiple device statuses and environmental parameters in Example 2. The fluctuation characteristics of the curves reflect the changes in device operating status and water quality environment collected by the industrial IoT gateway, which is beneficial for the confidence rule base to allocate confidence levels based on dissolved oxygen, temperature, and device status reference levels, supporting the effective operation of the multi-device collaborative decision-making model based on cooperative game theory.

[0054] Example 3:

[0055] In practical implementation, the game decision-making module obtains a parameterized feature vector from the confidence inference module. This parameterized feature vector is a 12-dimensional vector, with each dimension representing the confidence assignment values ​​for low dissolved oxygen level, medium dissolved oxygen level, high dissolved oxygen level, low temperature level, medium temperature level, high temperature level, normal state level of aerator, abnormal state level of aerator, normal state level of feeder, abnormal state level of feeder, normal state level of water pump, and abnormal state level of water pump. The game decision-making module defines the aerator, feeder, and water pump as different game participants in the cooperative game model, denoted as aerator game participant, feeder game participant, and water pump game participant, respectively. The private information types corresponding to the participants in the aerator game are defined as dimensions related to aerator operation and dissolved oxygen regulation in the parameterized feature vector, including confidence assignment values ​​for low dissolved oxygen level, medium dissolved oxygen level, high dissolved oxygen level, normal state level, and abnormal state level of the aerator. Similarly, the private information types corresponding to the participants in the feeder game are defined as dimensions related to feeder operation and temperature regulation in the parameterized feature vector, including confidence assignment values ​​for low temperature level, medium temperature level, high temperature level, normal state level, and abnormal state level of the feeder. Finally, the private information types corresponding to the participants in the water pump game are defined as dimensions related to water pump operation and flow regulation in the parameterized feature vector, including confidence assignment values ​​for normal state level, abnormal state level, all temperature level, and all dissolved oxygen level.

[0056] Strategy spaces are defined for the participants in the aerator, feeder, and water pump games, respectively. The strategy space for the aerator players consists of five discrete power output levels, corresponding to 0%, 25%, 50%, 75%, and 100% of the rated power, represented by the level values ​​0.00, 0.25, 0.50, 0.75, and 1.00, respectively. The strategy space for the feeder players also consists of five discrete power output levels, corresponding to 0%, 25%, 50%, 75%, and 100% of the rated power, represented by the level values ​​0.00, 0.25, 0.50, 0.75, and 1.00, respectively. The strategy space for the water pump players also consists of five discrete power output levels, corresponding to 0%, 25%, 50%, 75%, and 100% of the rated power, represented by the level values ​​0.00, 0.25, 0.50, 0.75, and 1.00, respectively. A payoff function is defined for each game participant, representing the difference between the water quality improvement benefit and the energy consumption cost obtained by that participant under the combined strategy of all game participants. In implementation, the game decision-making module pre-configures a water quality improvement benefit mapping table. The inputs to this table are the gear values ​​of the aerator, feeder, and water pump game participants, as well as the current parameterized feature vector. The output is the water quality improvement benefit value, a dimensionless scalar, calculated offline by a multivariate nonlinear regression model fitted from historical operating data and stored in the mapping table. The energy consumption cost is calculated separately according to the following formula:

[0057]

[0058] in, Indicates the first The energy consumption cost of each game participant A value of aer represents a participant in the aerator game. When the value is "feed", it represents a participant in the bait machine game. When the value is "pump", it represents a participant in the pump game. Indicates the first Energy cost coefficient of each game participant The price is 0.8 yuan per kilowatt-hour. Take 0.5 yuan per kilowatt-hour. The energy cost coefficient is set at 0.6 yuan per kilowatt-hour. The above energy cost coefficient is determined by multiplying the industrial and commercial electricity price in the region where the equipment is located by the equipment power factor. Indicates the first The rated power of each game participant It is 3 kilowatts. It is 1.5 kilowatts. It is 2 kilowatts; Indicates the first The power level value of each game participant under their current strategy. ; This represents the duration of the control command, fixed at 0.5 hours. The payoff function for the participants in the aerator game is: The payoff function for participants in the bait machine game is: The payoff function for the participants in the pump game is: ,in This refers to the water quality improvement benefit value obtained from the water quality improvement benefit mapping table.

[0059] The game decision-making module solves for the Nash equilibrium by iteratively executing the optimal response algorithm. Before the iteration begins, the game decision-making module randomly initializes the first initial power level of the aerator game participants. The second initial power level for participants in the bait machine game. And the third initial power level of the participants in the water pump game. The randomization method involves randomly selecting a value from the set {0.00, 0.25, 0.50, 0.75, 1.00} according to a uniform distribution, and combining these three initial power levels into the current strategy combination. .

[0060] In the first iteration, the game decision-making module fixed the second initial power level of the bait machine game participants. The third initial power level of the participants in the game with the water pump Keeping the power levels constant, enumerate all five selectable power levels for the participants in the aerator game, and select them sequentially. For each type of aerator, a power setting is selected, combined with a fixed value. and A temporary strategy combination is formed. The game decision module uses the temporary strategy combination and the current parameterized feature vector as an index to query the water quality improvement benefit mapping table, obtain the water quality improvement benefit value, and calculate the energy consumption cost of the aerator game participants according to the formula. Then, calculate the payoff function values ​​of the participants in the oxygenator game. The game-theoretic decision-making module compares the five selectable power levels. Value, select to make The highest power setting serves as a new strategy for participants in the oxygenator game. If multiple power levels exist... If the values ​​are tied for the largest, then the one with the smallest power level value is selected as the new strategy.

[0061] In the second iteration, the game decision module fixed the new strategies of the aerator game participants. The third initial power level of the participants in the game with the water pump Unchanged, enumerate all five selectable power levels for the bait machine game participants. Similarly, query the water quality improvement benefit mapping table and calculate the payoff function values ​​for the participants in the bait-feeding game. Select to make The highest power level is used as a new strategy for participants in the bait machine game. When parallel, take the lowest gear.

[0062] In the third iteration, the game decision module fixed the new strategies of the aerator game participants. New strategies for participants in the bait machine game Enumerate all five selectable power levels for the participants in the water pump game. Query the water quality improvement benefit mapping table and calculate the payoff function values ​​of the participants in the water pump game. Select to make The highest power setting is used as a new strategy for participants in the water pump game. When parallel, take the lowest gear.

[0063] After completing the above three rounds of iterations, the game decision-making module checks the newly generated strategy combinations. Does it combine with the previous strategy? They are exactly the same. If the power levels of the three game participants do not change, then stop the iteration and combine the strategies. This is taken as the Nash equilibrium solution. If the power level of any player changes, the current strategy combination is replaced with a new one, and the aforementioned three-round iterative process is repeated until the power levels of the aerator, feeder, and pump players no longer change in two consecutive rounds. The game decision module sets a maximum of 20 iterations. If convergence is not achieved after 20 rounds, the strategy combination obtained in the 20th iteration is taken as the Nash equilibrium solution.

[0064] The game decision-making module will use the discretized power output levels corresponding to the aerator game participants in the Nash equilibrium solution. This is converted into an aerator control command, which includes the device identification string "aerator" and the target power value. kilowatts and a duration of 1800 seconds; discrete power output levels corresponding to the participants in the bait machine game. This is converted into feeder control commands, which include the device identification string "feeder" and the target power value. kilowatts and a duration of 1800 seconds; the discretized power output levels corresponding to the participants in the water pump game. This is converted into a water pump control command, which includes the device identification string "pump" and the target power value. The power output is kilowatts and the duration is 1800 seconds. The above three control commands constitute the optimal control command set and are output to the command distribution module.

[0065] Example 4:

[0066] In practical implementation, the instruction distribution module obtains the optimal control instruction set from the game theory decision-making module. This optimal control instruction set includes three instructions: an aerator control instruction, a feeder control instruction, and a water pump control instruction. The aerator control instruction includes the device identifier string "aerator", the target power value, and a duration of 1800 seconds; the feeder control instruction includes the device identifier string "feeder", the target power value, and a duration of 1800 seconds; and the water pump control instruction includes the device identifier string "pump", the target power value, and a duration of 1800 seconds. The instruction distribution module extracts a first control instruction targeting the aerator actuator from the aerator control instruction. The specific content of the first control instruction is the target power value and duration from the aerator control instruction. The instruction distribution module encapsulates the first control instruction into a first message frame conforming to the aerator communication protocol. The aerator communication protocol adopts a proprietary application layer protocol based on the CAN bus extended frame format. The first message frame structure includes a 29-bit identifier field and an 8-byte data field. The high 5 bits of the identifier field are fixed at 0x06, representing the aerator node address. The first 2 bytes of the data field encode the target power value as a multiple of 1000 in big-endian order. The 3rd and 4th bytes of the data field encode the duration in seconds in big-endian order. The last 4 bytes of the data field are reserved as 0x00. The instruction distribution module extracts the second control instruction for the feeder actuator from the feeder control instructions and encapsulates the second control instruction into a second message frame conforming to the feeder communication protocol. The feeder communication protocol also adopts the CAN bus extended frame format. The high 5 bits of the identifier field of the second message frame are fixed at 0x07, representing the feeder node address, and the data field encoding method is the same as that of the first message frame. The instruction distribution module extracts the third control instruction for the water pump actuator from the water pump control instructions and encapsulates the third control instruction into a third message frame conforming to the water pump communication protocol. The high 5 bits of the water pump communication protocol identifier field are fixed at 0x08, representing the water pump node address. The data field is encoded in the same way as the first message frame.

[0067] The command distribution module appends the same global execution timestamp to the first, second, and third message frames. The global execution timestamp is generated as follows: the command distribution module reads the current time value of the IEEE 1588 precision time protocol clock maintained internally by the industrial IoT gateway, and adds a preset command advance time offset to the current time value. Get the global execution timestamp .in Calculate according to the following formula:

[0068]

[0069] In the formula, This represents the maximum communication delay from sending a message frame from the output channel of the industrial IoT gateway to successfully receiving the message frame at the corresponding actuator communication interface. The value is fixed at 2 milliseconds. This value is determined by taking the 99th percentile value after multiple measurements of the frame transmission time at the maximum bus load rate in the actual deployed CAN bus network. This indicates the maximum processing time required for the actuator communication interface to receive a message frame, parse the control command, and begin execution. The value is fixed at 1.5 milliseconds, which is obtained by rounding down the datasheet parameter of the lowest-performing of the three actuators: aerator, feeder, and water pump. This indicates the clock skew protection margin after the industrial IoT gateway synchronizes with other devices using the IEEE 1588 precise time protocol. The value is fixed at 500 microseconds, set based on the results of synchronization accuracy retention tests, to absorb clock drift during synchronization message exchange intervals. The calculated global execution timestamp... It is a 64-bit unsigned integer, with the high 32 bits representing the number of seconds since the IEEE 1588 epoch and the low 32 bits representing the number of nanoseconds. The instruction dispatch module will use the global execution timestamp. The 64-bit value is appended to the end extension field of the first message frame. The extension field is located after the CAN data field and before the end-of-frame bit, occupying 8 bytes; similarly, the identical global execution timestamp is appended. The 64-bit value is appended to the extended field of the second message frame and the extended field of the third message frame. The command distribution module simultaneously pushes the first, second, and third message frames, with the global execution timestamp appended, to the command sending queue of the industrial IoT gateway. The command sending queue is a priority message queue maintained by the real-time operating system kernel, with a message depth of 32 messages, and each message stores a complete CAN message frame data.

[0070] In the real-time operating system scheduler of the industrial IoT gateway, a synchronous transmission task is created, and its priority is set to the highest real-time priority in the real-time operating system to ensure that it is not preempted by other application tasks. The synchronous transmission task reads the first message frame with a globally timestamped execution date from the command transmission queue and temporarily stores it in a local variable within its task stack. Then, the synchronous transmission task continues to read the second and third message frames with globally timestamped execution dates from the command transmission queue, temporarily storing them in two other local variables within its task stack, respectively. The reading operation is performed in a non-blocking loop. If the command transmission queue is empty during each reading attempt, it yields for 1 microsecond and retryes, until all three complete message frames have been read.

[0071] The synchronous transmission task sets its own state to suspended, calls the task suspension primitive provided by the real-time operating system scheduler, causing the scheduler to remove the synchronous transmission task from the ready task list and immediately trigger a task context switch to the second-highest priority task. The real-time operating system scheduler periodically checks whether the current system clock has reached the globally executed timestamp, with the check period set to 50 microseconds. In each system clock tick interrupt service routine, the real-time operating system scheduler compares the current nanosecond count value of the IEEE 1588 synchronous clock with the globally executed timestamp. The difference between the nanosecond values. When the difference is first less than or equal to 50 microseconds, the real-time operating system scheduler enters a precision time wait state: all maskable interrupts are disabled, and a single-trigger compare interrupt is set using a hardware timer, with the compare value being... The nanosecond field value. When a hardware timer compare match event occurs, the hardware timer compare interrupt service routine is triggered.

[0072] In the hardware timer compare interrupt service routine, the real-time operating system scheduler simultaneously wakes up the first coroutine associated with the aerator output channel, the second coroutine associated with the feeder output channel, and the third coroutine associated with the water pump output channel. The first, second, and third coroutines are all lightweight cooperative tasks managed by the real-time operating system scheduler, each bound to a CAN controller transmit mailbox. The wake-up process is implemented using a coroutine recovery primitive provided by the real-time operating system scheduler. This primitive clears the suspend flag in the target coroutine's control block and adds the target coroutine to the head of the scheduler's run queue, scheduling its execution immediately in priority order after the interrupt service routine exits. Since the three coroutines have the same priority and are added to the head of the run queue simultaneously, the real-time operating system scheduler executes the first, second, and third coroutines sequentially according to the run queue order.

[0073] After the first coroutine gains execution rights, it drives the first transmit mailbox of the CAN controller associated with the aerator output channel, writes the first message frame into the frame data register of the first transmit mailbox of the CAN controller, and sets the transmit request bit. When the CAN controller detects that the bus is idle, it transmits the first message frame on the physical CAN bus corresponding to the first output channel. The second coroutine gains execution rights in the next time slot of the run queue, drives the second transmit mailbox of the CAN controller associated with the feeder output channel, writes the second message frame into the frame data register of the second transmit mailbox, sets the transmit request bit, and transmits the second message frame on the second output channel. The third coroutine gains execution rights in the next time slot after the run queue, drives the third transmit mailbox of the CAN controller associated with the water pump output channel, writes the third message frame into the frame data register of the third transmit mailbox, sets the transmit request bit, and transmits the third message frame on the third output channel. The execution interval of the three coroutines is determined only by the coroutine switching instruction cycle, controlled on the order of microseconds, ensuring that the start time of transmission of the first, second, and third message frames on their respective output channels is relative to the global execution timestamp. The time deviation is no more than 1 millisecond.

[0074] Example 5:

[0075] In practical implementation, the feedback update module starts operating after the instruction distribution module sends differentiated execution instructions synchronized with timestamps to the aerator, feeder, and water pump via the industrial IoT gateway. The feedback update module receives execution completion confirmation frames from the aerator, feeder, and water pump actuators respectively through the data aggregation layer of the industrial IoT gateway. The aerator execution completion confirmation frame is a CAN bus extended frame conforming to the aerator communication protocol. The high 5 bits of the frame identifier are 0x06, the first byte of the data field is the confirmation frame type code 0xA1, the second to fifth bytes of the data field encode the 32-bit UNIX timestamp integer value of the actual execution time in big-endian order, the sixth and seventh bytes of the data field encode the integer value of the actual execution power value as a multiple of 100 in big-endian order, and the eighth byte of the data field is reserved as 0x00. The feeder execution completion confirmation frame has a frame identifier of 0x07, a frame type code of 0xA1, and the data field encoding method is the same as that of the aerator execution completion confirmation frame. The high 5 bits of the water pump completion confirmation frame identifier are 0x08, the frame type code is 0xA1, and the data field encoding method is the same as that of the aerator completion confirmation frame. After receiving the above completion confirmation frame, the data aggregation layer of the industrial IoT gateway forwards the raw data of the completion confirmation frame frame by frame to the internal buffer of the feedback update module.

[0076] The feedback update module extracts the UNIX timestamp integer value of the actual execution time from bytes 2 to 5 of the data field of the aerator completion confirmation frame. The UNIX timestamp integer value is converted into a floating-point value in seconds representing the actual execution time. The actual execution power value of the aerator is obtained by extracting a multiplied integer value (in multiples of 100) from bytes 6 and 7 of the data field of the aerator completion confirmation frame and dividing it by 100.0. The unit is kilowatts. The feedback update module extracts the actual execution time value of the feeder from the feeder completion confirmation frame in the same way. Actual power output of the feeder The feedback update module extracts the actual execution time value of the water pump from the water pump completion confirmation frame. and the actual power value of the water pump .

[0077] The feedback update module reads the expected execution time and target power value from the command sending history of the most recently issued differentiated execution command from the command distribution module. The expected execution time of the aerator's differentiated execution command is denoted as... The target power value is denoted as The expected execution time of the differentiated execution instructions of the feeder is denoted as . The target power value is denoted as The expected execution time of the differentiated execution command for the water pump is denoted as . The target power value is denoted as .

[0078] The feedback update module will update the actual execution time of the aerator. With the expected execution time of the aerator Compare and calculate the aerator execution time deviation. The actual execution time of the feeder With the expected execution time of the feeder Compare and calculate the execution time deviation of the feeder. ; the actual execution time of the water pump With the expected execution time of the water pump Compare and calculate the pump execution time deviation. The feedback update module will update the actual operating power value of the aerator. With the target power value of the aerator Compare and calculate the deviation in the aerator's operating power. The actual power value of the feeder With the target power value of the feeder Compare and calculate the deviation of the feeder's execution power. The actual power value of the water pump With the target power value of the water pump Compare and calculate the deviation in pump operating power. .

[0079] The feedback update module packages the six calculated deviation values ​​into a single feedback signal vector. and the feedback signal vector The input is fed into the parameter updater of a multi-device collaborative decision-making model built on cooperative game theory. The parameter updater is an independent functional unit within the multi-device collaborative decision-making model. Its core architecture consists of a closed-loop correction circuit composed of a proportional-integral (PI) regulator and a saturation limiter connected in series. The PI regulator comprises two parallel branches: a proportional branch and an integral branch. The proportional branch amplifies the execution power deviation component in the feedback signal vector using proportional gain, while the integral branch amplifies the accumulated value of the execution time deviation component in the feedback signal vector using integral gain. The sum of the outputs of the two branches is then fed into the saturation limiter.

[0080] The parameter updater adjusts the energy cost coefficient in the payoff function of each participant in the multi-device collaborative decision-making model based on the feedback signal vector. The parameter updater reads the current energy cost coefficients of the aerator participants. And calculate the updated energy cost coefficients of the aerator game participants according to the following formula. :

[0081]

[0082] In the formula, The integral gain constant representing the update of the energy cost coefficient has a fixed value. (This value was determined by conducting a step test on the power deviation response of the aerator during the system debugging phase, selecting a gain value that makes the overshoot less than 10% and the settling time less than 3 update cycles.) A continuous function representing the deviation of the aerator's execution power within the integral time window (unit: ); This indicates the execution cycle (in seconds) of the feedback update module. The integral term is discretized using the rectangular approximation method, which extracts multiple terms sampled within the integration interval. The summation of the values ​​multiplied by the sampling interval is an approximation.

[0083] The parameter updater updates the energy cost coefficients of the bait machine game participants using the same formula structure. Energy cost coefficient of participants in the game with water pumps Update the corresponding execution power deviation or and the corresponding execution time deviation or Proportional gain constant and integration time constant The same value is applied to all three game participants.

[0084] The updated energy cost coefficient is then limited by a saturation limiter. The saturation limiter applies the updated energy cost coefficients of the aerator game participants. Perform boundary checks, if If the price is below the lower limit of 0.3 yuan per kilowatt-hour, it will be forcibly set to 0.3 yuan. If the price exceeds the upper limit of 2.0 yuan per kilowatt-hour, it will be forcibly set to 2.0. The lower limit of 0.3 yuan per kilowatt-hour is determined based on the minimum residential electricity price of the power grid, and the upper limit of 2.0 yuan per kilowatt-hour is set based on the maximum allowable multiple of peak-valley electricity price for the unit energy consumption cost of equipment in the system. Energy consumption cost coefficient of participants in the baiting machine game. The lower limit is set at 0.2 yuan per kilowatt-hour, and the upper limit is set at 1.5 yuan per kilowatt-hour; the energy consumption cost coefficient of the participants in the water pump game. The lower limit is set at 0.25 yuan per kilowatt-hour, and the upper limit is set at 1.8 yuan per kilowatt-hour. The three energy cost coefficients after the limiting process are written to the non-volatile storage area inside the parameter updater, overwriting the old values. When the game decision module calculates the optimal control instruction set again, the parameter updater will store the latest energy cost coefficients of the aerator game participants. Energy cost coefficient of participants in the bait machine game Energy cost coefficient of participants in the game with water pumps Provided to the game decision-making module, it is used to calculate the energy consumption costs of the aerator game participants, feeder game participants, and water pump game participants.

[0085] See Figure 6 The horizontal axis in the graph represents the number of times the energy cost coefficient of the feedback update module is updated. The vertical axis represents the energy cost coefficient of each game participant, in yuan / kWh. The blue solid line represents the energy cost coefficient of the participants in the aerator game. The orange dashed line represents the energy cost coefficient of the participants in the bait machine game. The green dotted line represents the energy cost coefficient of the participants in the water pump game. .

[0086] With the number of updates With the increase in [something], the three curves generally show a gradual adjustment trend. [Aerator energy consumption cost coefficient] Starting from an initial value of 0.8 yuan / kWh, the cost showed a slow upward trend in the first 100 updates, then reached a local peak of nearly 0.83 yuan / kWh around the 200th update, before gradually declining and stabilizing around 0.77 yuan / kWh. (Feeder energy consumption cost coefficient) Starting from an initial value of 0.5 yuan / kWh, the price showed a steady and gradual downward trend, accelerating particularly after 400 updates, eventually approaching 0.42 yuan / kWh. (Water pump energy cost coefficient) Starting from an initial value of 0.6 yuan / kWh, the curve fluctuates more than that of aerators and feeders, showing periodic fluctuations. The range of variation is roughly between 0.58 and 0.65 yuan / kWh, and the overall trend shows a slight downward trend after 300 updates.

[0087] The curves show that the feedback update module dynamically adjusts the energy cost coefficients of each game participant based on the actual execution power and time deviation feedback. Specifically, the aerator's energy cost coefficient adjustment primarily exhibits a process of first increasing and then decreasing, indicating that the system meticulously corrects the aerator's cost weight to balance power and time deviations. The feeder's energy cost coefficient shows a continuous decrease, indicating that the system gradually reduces the feeder's cost weight, possibly corresponding to feedback adjustment as the feeder's execution power deviation tends towards a negative value. The water pump's energy cost coefficient fluctuates significantly, reflecting the alternating influence of execution time and power deviations during the control process, demonstrating the feedback update module's dynamic adaptability to water pump cost parameters.

[0088] Overall, none of the three curves exceeded the upper and lower limits set by the saturation limiter (0.3~2.0 yuan / kWh for aerator, 0.2~1.5 yuan / kWh for feeder, and 0.25~1.8 yuan / kWh for water pump), indicating that the parameter update process is stable and effective, and can reasonably adjust the energy consumption cost coefficients of the game participants based on feedback signals, thus promoting the optimized operation of the multi-device collaborative decision-making model.

[0089] The above description is merely a specific embodiment of this application, but the scope of protection of this application 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 this application should be included within the scope of protection of this application.

Claims

1. A multi-equipment collaborative control system for aquaculture based on the Industrial Internet of Things, characterized in that, include: The data acquisition module is used to respond to the heterogeneous device data streams aggregated by the industrial IoT gateway and construct a multimodal time-series dataset containing dissolved oxygen, temperature and device status parameters. The time-domain alignment module is used to generate synchronized data frames with a unified timestamp reference by performing a sliding window-based time-domain alignment operation on the multimodal time-series dataset. The confidence reasoning module is used to perform uncertainty quantification processing on each data point in the synchronous data frame using a confidence rule base based on evidence reasoning, so as to obtain a parameterized feature vector with confidence distribution. The game decision module is used to input the parameterized feature vector into a multi-device collaborative decision model based on cooperative game theory, and calculate the optimal control instruction set that makes the comprehensive energy consumption and water quality control error of multiple aquaculture equipment reach Nash equilibrium. The instruction distribution module is used to issue differentiated execution instructions synchronized with timestamps to the aerator, feeder and water pump respectively through the industrial Internet of Things gateway according to the optimal control instruction set.

2. The multi-equipment collaborative control system for aquaculture based on the Industrial Internet of Things as described in claim 1, characterized in that, In response to the heterogeneous device data streams aggregated by the industrial IoT gateway, a multimodal time-series dataset containing dissolved oxygen, temperature, and device status parameters is constructed, specifically including: The data aggregation layer of the industrial IoT gateway subscribes to the current signal of the aerator, the speed signal of the feeder, and the vibration signal of the water pump, and performs protocol-independent raw data capture operations on the current signal, the speed signal of the feeder, and the vibration signal of the water pump. The analog signals in the captured heterogeneous device data stream are converted into digital signals by an analog-to-digital converter, and a timestamp tag for the time of reception is attached to the digital signals. Based on the dissolved oxygen sensor, temperature sensor, and the data source identifier corresponding to each device, the digital signals with added timestamp tags are classified and stored in the circular buffer of the in-memory database to form a multimodal time series dataset.

3. The multi-equipment collaborative control system for aquaculture based on the Industrial Internet of Things as described in claim 2, characterized in that, By performing a sliding window-based time-domain alignment operation on the multimodal time-series dataset, a synchronized data frame with a unified timestamp reference is generated, specifically including: Read a multimodal time series dataset of a preset duration from the circular buffer of the memory database, and identify the starting data point with the earliest timestamp label in the multimodal time series dataset; Using the starting data point as a reference, slide backward according to a fixed window step size, and at each sliding window position, extract all data points that intersect with the window start time and end time; Perform linear interpolation on all data points under the same timestamp label extracted within each sliding window to ensure that the number of data points output by each sliding window is consistent, and package all data points associated with that sliding window into a synchronization data frame.

4. The multi-equipment collaborative control system for aquaculture based on the Industrial Internet of Things as described in claim 1, characterized in that, Uncertainty quantification is performed on each data point in the synchronized data frame using a confidence rule base based on evidence reasoning to obtain a parameterized feature vector with a confidence distribution, specifically including: The dissolved oxygen values ​​in the synchronized data frames are matched with multiple preset dissolved oxygen reference levels in the confidence rule base, and an initial confidence level assignment value is calculated for each dissolved oxygen reference level. The temperature values ​​in the synchronized data frames are matched with multiple preset temperature reference levels in the confidence rule base, and an initial confidence level assignment value is calculated for each temperature reference level. The device status parameters in the synchronized data frame are matched with the preset normal status level and abnormal status level in the confidence rule base, and the initial confidence level assignment value is calculated for the normal status level and abnormal status level respectively. All the calculated initial confidence scores are vectorized and concatenated according to their respective synchronous data frames to generate a parameterized feature vector in which each dimension corresponds to a reference level and the sum of the confidence scores between dimensions is one.

5. The multi-equipment collaborative control system for aquaculture based on the Industrial Internet of Things as described in claim 4, characterized in that, The dissolved oxygen values ​​in the synchronized data frames are matched with multiple preset dissolved oxygen reference levels in the confidence rule base, and an initial confidence score assignment value is calculated for each dissolved oxygen reference level, specifically including: Obtain the dissolved oxygen value in the synchronized data frame, and compare the dissolved oxygen value with the threshold range of three reference levels (low dissolved oxygen level, medium dissolved oxygen level, and high dissolved oxygen level) in the confidence rule base; When the dissolved oxygen value is within the threshold range of the low dissolved oxygen level, the confidence assignment value of the low dissolved oxygen level is set to one, while the confidence assignment values ​​of the medium and high dissolved oxygen levels are set to zero. When the dissolved oxygen value is located in the overlapping part of the threshold interval of two adjacent reference levels, the confidence assignment values ​​for the low dissolved oxygen level and the medium dissolved oxygen level are calculated according to the ratio of the distance of the dissolved oxygen value to the center value of the two reference levels, and the confidence assignment value for the high dissolved oxygen level is set to zero.

6. The multi-equipment collaborative control system for aquaculture based on the Industrial Internet of Things as described in claim 1, characterized in that, The parameterized feature vectors are input into a multi-device collaborative decision-making model based on cooperative game theory to calculate the optimal control instruction set that enables the comprehensive energy consumption and water quality control error of multiple aquaculture devices to reach Nash equilibrium. Specifically, this includes: Aerators, feeders, and water pumps are defined as different players in the cooperative game model, and the confidence distribution in the parameterized feature vector corresponding to each player is used as the private information type of that player. A strategy space is defined for each game participant, which contains multiple discretized power output levels for that game participant. A payoff function is defined for each game participant, which is used to calculate the difference between the water quality improvement benefit and the energy consumption cost based on the strategy combination jointly adopted by all game participants. By iteratively executing the optimal response algorithm, each player in the game chooses the strategy that maximizes their own payoff function, given the current strategies of other players, until all players no longer change their strategies. The resulting strategy combination is then taken as the Nash equilibrium solution. The discrete power output level corresponding to each game participant in the Nash equilibrium solution is converted into a control command containing the device identifier, target power value and duration, and all control commands are combined into an optimal control command set.

7. The multi-equipment collaborative control system for aquaculture based on the Industrial Internet of Things as described in claim 6, characterized in that, By iteratively executing the optimal response algorithm, each player, given the current strategies of other players, chooses the strategy that maximizes their own payoff function until all players no longer change their strategies. The resulting strategy combination is taken as the Nash equilibrium solution, which includes: The first initial power setting of the aerator, the second initial power setting of the feeder, and the third initial power setting of the water pump are randomly initialized, and the first initial power setting, the second initial power setting of the feeder, and the third initial power setting of the water pump are combined into the current strategy combination. In the first iteration, the second initial power setting of the feeder and the third initial power setting of the water pump are fixed. All possible power settings of the aerator are enumerated, the aerator revenue function value under each power setting combination is calculated, and the power setting that maximizes the aerator revenue function value is selected as the new strategy for the aerator. In the second iteration, the new strategy of fixing the aerator and the third initial power level of the water pump is used. All possible power levels of the feeder are enumerated, the feeder revenue function value under each power level combination is calculated, and the power level that maximizes the feeder revenue function value is selected as the new strategy of the feeder. In the third iteration, the new strategies for fixed aerators and feeders are enumerated. All possible power levels of the water pump are enumerated, the water pump revenue function value under each power level combination is calculated, and the power level that maximizes the water pump revenue function value is selected as the new strategy for the water pump. Repeat the above three-round iterative process until the power levels of all game participants no longer change in two consecutive rounds of iteration. The strategy combination obtained in the last iteration is then taken as the Nash equilibrium solution.

8. The multi-equipment collaborative control system for aquaculture based on the Industrial Internet of Things as described in claim 1, characterized in that, Based on the optimal control instruction set, differentiated execution instructions synchronized with timestamps are issued to the aerator, feeder, and water pump respectively through the industrial IoT gateway, specifically including: The first control instruction for the aerator is extracted from the optimal control instruction set, and the first control instruction is encapsulated into a first message frame conforming to the aerator communication protocol. At the same time, the second control instruction for the feeder and the third control instruction for the water pump are extracted from the optimal control instruction set, and encapsulated into a second message frame and a third message frame, respectively. The same global execution timestamp is appended to the first message frame, the second message frame, and the third message frame, and the three message frames with the appended global execution timestamp are simultaneously pushed to the command sending queue of the industrial IoT gateway. Through multiple parallel output channels of the industrial IoT gateway, at the time indicated by the global execution timestamp, the first message frame, the second message frame, and the third message frame are simultaneously sent to the aerator, the feeder, and the water pump, respectively.

9. The multi-equipment collaborative control system for aquaculture based on the Industrial Internet of Things as described in claim 8, characterized in that, The global execution timestamp uses network time synchronized based on the IEEE 1588 precise time protocol to ensure that the time synchronization error between multiple parallel output channels is less than 1 millisecond.

10. The multi-equipment collaborative control system for aquaculture based on the Industrial Internet of Things as described in claim 8, characterized in that, Through multiple parallel output channels of the industrial IoT gateway, at the time indicated by the global execution timestamp, a first message frame, a second message frame, and a third message frame are simultaneously sent to the aerator, feeder, and water pump, respectively, specifically including: Create a synchronous sending task in the real-time operating system scheduler of the industrial IoT gateway, and read the first, second, and third message frames with global execution timestamps attached from the command sending queue; The synchronous sending task is suspended, and the real-time operating system scheduler periodically checks whether the current system clock has reached the time indicated by the global execution timestamp. When the current system clock reaches the time indicated by the global execution timestamp, the real-time operating system scheduler simultaneously wakes up the first coroutine of the associated aerator output channel, the second coroutine of the associated feeder output channel, and the third coroutine of the associated water pump output channel. The first coroutine drives the first output channel to send the first message frame, the second coroutine drives the second output channel to send the second message frame, and the third coroutine drives the third output channel to send the third message frame.

Images