Intelligent acousto-optic joint dispersing system based on biological safety threshold regulation and control

Abstract

The invention discloses an intelligent acousto-optic joint dispersing system based on biological safety threshold regulation and control, and relates to the technical field of crossing of biological safety and intelligent ecological regulation and control. The system comprises a regional ecological characteristic quantification module which generates a non-linear threshold distribution sequence through an improved whale optimization algorithm, and adapts to ecological characteristics of an early warning area, a warning area, a repelling area and an observation area; the distributed sensor network realizes accurate acquisition and noise reduction of biological data; a biological data multi-scale noise reduction module dynamically optimizes a threshold value; the biological invasion risk prediction module analyzes the invasion probability and verifies the ecological bearing compliance; the intelligent acousto-optic regulation and control module is based on a PID neural network and a sliding mode control fusion algorithm, and combines model prediction control to realize acousto-optic power dynamic adjustment and regional linkage in a multi-constraint scene. According to the system, through multi-module cooperation, the real-time response capability and ecological adaptability of biological activities are improved, the scientificity and stability of the dispersing effect are enhanced, the misjudgment risk is reduced, and ecological safety and efficient operation of the system are guaranteed.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of biological safety and intelligent ecological regulation, and particularly relates to an intelligent sound-light combined dispersing system based on biological safety threshold regulation. BACKGROUND

[0002] With globalization and changes in the ecological environment, biological invasion (expansion of alien species, overpopulation of local harmful organisms) has become a key problem that threatens ecological balance, agricultural safety and biodiversity. Traditional prevention and control methods (spraying of chemical agents, manual driving) have drawbacks such as environmental pollution, low efficiency, poor targeting, and easy injury to beneficial organisms, and are difficult to meet the modern ecological safety demand of precise prevention and control and ecological friendliness.

[0003] Currently, some biological monitoring and dispersing systems based on sensors and sound-light technology have significant deficiencies in threshold adaptability, data reliability, risk prediction accuracy, and regulation scientificity due to technical design flaws, and cannot meet the high-precision prevention and control demand of complex ecological scenes. The specific problems are as follows: Insufficient biological safety threshold adaptability and dynamics: Most systems use fixed or linear thresholds without considering regional ecological sensitivity index (ESI) differences, resulting in high risk of misjudgment in high sensitivity areas and missed judgment in low sensitivity areas; traditional optimization algorithms (such as basic whale optimization algorithm) are prone to local optimum and difficult to generate nonlinear threshold sequences that adapt to multiple regions (warning zone, alert zone, driving zone, observation zone), and the threshold does not match the biological activity pattern, resulting in system response lag.

[0004] Poor sensor network and data collection accuracy: Sensors are usually evenly distributed without nonlinear adjustment of node spacing according to regional importance, resulting in monitoring blind spots and resource waste; traditional Kalman filtering is prone to numerical overflow due to covariance matrix problems, and lacks EMI suppression circuit, resulting in large biological data errors and low reliability, which directly affects the accuracy of risk assessment.

[0005] Data denoising and threshold optimization lag: Traditional denoising techniques (such as wavelet transform) rely on subjective wavelet bases and are difficult to adapt to biological data (nonlinear, non-stationary) characteristics, and are prone to lose key features; threshold adjustment relies on manual or simple comparison without combining time series prediction models (such as LSTM-HMM), which is prone to response lag or over-response.

[0006] Weak risk prediction and ecological carrying capacity verification: Traditional HMM model is difficult to capture long-term characteristics of biological activity, resulting in large risk prediction errors; finite element method is inefficient in solving biological diffusion equation, and cannot detect ecological carrying capacity overrun in real time, missing the prevention and control window period.

[0007] Poor sound-light regulation and linkage effect: traditional PID control parameters are fixed, difficult to adapt to environmental changes, easy to appear power chattering, leading to incomplete dispersal or misinjury of beneficial organisms; lack of multi-region collaborative optimization mechanism, adjacent areas respond out of sync, leaving dispersal loopholes.

[0008] In summary, the prior art cannot meet the high-precision prevention and control needs of complex ecological scenarios, and the present application proposes an intelligent sound-light joint dispersal system based on biological safety threshold regulation. SUMMARY

[0009] In order to overcome the shortcomings and deficiencies of the prior art, the present application adopts the following technical solutions: In summary, due to the adoption of the above technical solutions, the present application has the following beneficial effects: 1. The present application generates a nonlinear threshold sequence by improving the whale optimization algorithm TIWOA combined with the ecological sensitivity index ESI, accurately adapts to the ecological characteristics of the warning area, the alert area, the drive area, and the observation area, solves the problem of disconnection between traditional fixed / linear threshold and ecological characteristics; at the same time, relying on the Legendre polynomial node layout, MSF-EKF filtering, EMI static noise circuit and VMD+EMD+self-encoder multi-scale noise reduction technology, the present application realizes accurate collection and noise suppression of biological data including species, quantity and trajectory, greatly reduces the probability of misdriving beneficial organisms and missing invasion risks, and lays a reliable data foundation for subsequent risk assessment.

[0010] 2. The present application uses a LSTM-HMM hybrid model to capture long-term dependence of biological activity, improving the timing accuracy of invasion risk probability prediction; combined with the physical information neural network PINN to replace the traditional finite element method to solve the biological diffusion equation, and equipped with adaptive grid encryption technology, the present application realizes efficient calculation of biological density spatio-temporal distribution and real-time verification of ecological carrying compliance, can detect local biological aggregation anomalies in advance, avoid missing the prevention and control window period, and effectively protect the safety of the ecological system.

[0011] 3. The present application is based on the PIDNN-sliding mode-MPC fusion control strategy, dynamically optimizes the output power of sound-light equipment, balances the dispersal effect and ecological protection, and avoids misinjury of beneficial organisms; at the same time, through MPC processing of multi-constraint scenarios, the present application realizes the timing linkage of multi-region sound-light equipment, solves the dispersal loopholes caused by the out-of-sync response of traditional systems, and improves the response speed and operation stability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creating any inventive labor.

[0013] Figure 1 The module diagram of the intelligent sound and light combined dispersion system based on biosafety threshold regulation of the application is shown. Figure 2 The working flow chart of the system of the application is shown. Figure 3 The flow chart of the region division and biosafety threshold setting of the application is shown. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0015] In addition, the described features, structures or characteristics can be combined in any suitable manner in one or more example embodiments. In the following description, many specific details are provided to give a sufficient understanding of example embodiments of the disclosure. However, a person of ordinary skill in the art will realize that the technical solutions of the disclosure can be practiced without one or more of the specific details, or other methods, components, steps, etc. can be used. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring aspects of the disclosure.

[0016] Embodiment 1: Referring to Figure 1 As shown, the intelligent sound and light combined dispersion system based on biosafety threshold regulation of the embodiment includes a region threshold setting module, a sensor data acquisition module, a data processing optimization module, a risk verification module and a sound and light regulation module.

[0017] The region threshold setting module: based on the ESI quantification results, a non-linear threshold distribution sequence is generated by improving the whale optimization algorithm to adapt to the ecological characteristics of the warning area, the alert area, the dispersal area and the observation area, and to realize the global optimization and dynamic adjustment of the threshold.

[0018] The sensor data acquisition module: the main sensor is arranged according to the Legendre polynomial node distribution rule, and the real-time biological data (species, quantity, activity track) is acquired by combining the redundant sensor, and the data error correction and noise suppression are performed by the MSF-EKF filter and the EMI static noise circuit.

[0019] Data processing optimization module: adopt VMD+EMD double decomposition algorithm and self-encoder reconstruction technology to process biological data with multi-scale noise reduction, combine fuzzy logic algorithm and LSTM-HMM hybrid model to dynamically optimize biological safety threshold range, generate risk deviation signal and regulation instruction.

[0020] Risk verification module: analyze biological intrusion probability through LSTM-HMM hybrid model, combine PINN to solve biological diffusion equation and adaptive grid encryption technology to verify ecological carrying compliance, trigger early warning signal and generate regulation instruction.

[0021] Acousto-optic regulation module: based on PID neural network (PIDNN) and sliding mode control fusion algorithm, combine model predictive control (MPC) to optimize acousto-optic power adjustment and regional linkage effect, realize dynamic regulation and equipment working mode adaptation in multi-constraint scene.

[0022] The beneficial effects of the embodiment are: through multi-module cooperation, threshold dynamic optimization, accurate data acquisition and noise reduction, risk prediction and ecological verification integration are realized, the real-time response capability and ecological adaptability of the system to biological activities are improved, the scientificity and stability of the dispersing effect are enhanced, the misjudgment risk is reduced, and the ecological safety and efficient operation of the system are ensured.

[0023] Embodiment 2: Referring to Figure 2 The intelligent acousto-optic combined dispersing system based on biological safety threshold regulation of the embodiment has the following working process: Step one, regional division and biological safety threshold setting.

[0024] The monitoring area is divided into early warning area (frequent biological activity but not intrusion), alert area (potential intrusion risk), drive-off area (high intrusion risk) and observation area (ecologically sensitive core area), each area is controlled by independent acousto-optic unit output power, and biological activity state is monitored in real time by biological sensor; The preset biological safety threshold range of each area is a nonlinear distribution sequence; the threshold range setting steps include: introducing improved whale optimization algorithm TIWOA combined with ecological sensitivity index, realizing global optimization of threshold interval through nonlinear convergence factor and sinusoidal spiral update strategy.

[0025] Referring to Figure 3 The specific process of regional division and biological safety threshold setting is as follows: S11, ecological characteristic quantification: Quantify the ecological sensitivity index ESI of each area through field investigation, remote sensing data or ecological model (such as InVEST model), for example: wetland ESI=0.9 (high sensitivity), farmland ESI=0.3 (low sensitivity), and urban fringe ESI=0.5 (medium sensitivity).

[0026] S12, design a sequence of nonlinear threshold distribution: the threshold range is dynamically adjusted according to ESI. For example, the pre-warning zone threshold range is [0.8, 1.0] (allow normal activities), the alert zone threshold range is [0.4, 0.7] (trigger potential risk response), the evacuation zone threshold range is [0.4, 0.7] (quickly trigger dispersion), and the observation zone threshold range is [0.1, 0.3] (strictly limit biological activities).

[0027] S13, improve the parameter design of whale optimization algorithm TIWOA: TIWOA introduces a nonlinear function (such as an exponential decay or a piecewise function) to dynamically adjust the search step size: ; wherein, is the convergence factor of the tth iteration, the initial value (the conventional initial value of the whale optimization algorithm, suitable for global search requirements); t is the current iteration number, the value range ; is the maximum number of iterations, set according to the dimension of the threshold combination, when optimizing the upper and lower limits of the thresholds of the four regions (pre-warning zone, alert zone, evacuation zone, and observation zone), (balance optimization precision and efficiency); k is the exponential decay coefficient, which ensures initial large step size global search and later small step size local fine optimization. The value of the exponential decay coefficient k is positively correlated with the complexity of the regional ecological sensitivity: when the monitoring region contains high sensitivity (such as wetland ESI=0.9) and low sensitivity (such as farmland ESI=0.3) mixed regions, k takes 0.04-0.05 to enhance the local optimization precision; when the sensitivity of the region is uniform (such as a single farmland region), k takes 0.01-0.02 to speed up the convergence speed.

[0028] The sinusoidal spiral strategy is used to enhance population diversity and avoid local optimum, and the spiral angle needs to change with the iteration period to balance the exploration and development ability, the specific design is as follows: Spiral angle calculation formula: ; The position update formula is: ; wherein, is the spiral angle of the tth iteration, the value range [−1.0, 1.0] (periodic change to realize the contraction and expansion of the spiral trajectory); is the angle amplitude, the value is 0.8-1.2 (default 1.0, suitable for the nonlinear distribution characteristics of biological thresholds); is the angle change period, the value (for example, , , ensure that the angle completes 3 cycles of change in the iteration process); Initial phase, value 0 (simplified calculation, does not affect the diversity of spiral trajectory); : Distance vector between current individual and optimal individual; l is the spiral step vector, value 0.5-1.0 (default 0.8, control the winding density of spiral trajectory).

[0029] Parameter adaptation rule: when the population falls into local optimum (the change of fitness function value is less than 10−4 for 10 consecutive generations), temporarily increase to 1.2 to enhance the diversity of the trajectory to escape from the local trap; in the later iteration period ( ), reduce l to 0.5 to narrow the spiral range to fine-tune the threshold.

[0030] The fitness function needs to balance the three major goals of "threshold-ESI matching degree, threshold nonlinearity, and system response efficiency". The specific quantitative design is as follows: The complete expression of the fitness function is: ; Among them, is the threshold-ESI matching degree weight; is the threshold nonlinearity characteristic weight; is the system response efficiency weight; is the Pearson correlation coefficient; is the Gini coefficient. In this fitness function, the essence of "threshold nonlinearity characteristic" is that the distribution of threshold is not linear and uniform, but there is an uneven nonlinear feature. Therefore, using the "unbalancedness measurement" logic of Gini coefficient, the "unbalancedness degree of threshold-related data (such as threshold values in different monitoring scenarios and different positions)" is quantified by Gini coefficient G: the greater G is, the more uneven the threshold distribution is, and the more significant the "threshold nonlinearity characteristic" is; the smaller G is, the more balanced the threshold distribution is, and the weaker the "threshold nonlinearity characteristic" is; is the response delay coefficient, , is the average response delay, is the target delay.

[0031] Weight coefficient adjustment mechanism: when the monitoring area is the ecological sensitive core area (such as observation area ESI=0.1-0.3), increase to 0.6, decrease to 0.1; when the area is a non-sensitive area (such as farmland), increase to 0.3, decrease to 0.4.

[0032] The penalty function is used to integrate the "high ESI area threshold upper and lower limit constraint" into the fitness function to ensure that the threshold setting meets the ecological safety bottom line. The specific design is as follows: Penalty function expression: ; wherein P(Th) is the penalty term value, P(Th) = 0 when the threshold meets the constraint, and increases with the deviation when it exceeds the constraint; μ is the penalty coefficient, taking values 5.0-10.0 (10.0 for high ESI area, 8.0 for medium ESI area, 5.0 for low ESI area, reinforcement sensitive area constraint); , is the current optimized area threshold lower limit, upper limit; , is the constraint value of the area threshold lower limit, upper limit (according to ESI setting: high ESI area (ESI ≥ 0.7) , ; medium ESI area (0.3 < ESI < 0.7) , ; low ESI area (ESI ≤ 0.3) , ).

[0033] Fitness function with fusion penalty term: the final fitness function is the sum of the original function and the penalty term, that is , to ensure that the threshold constraint of the high ESI area is met first in the optimization process.

[0034] TIWOA global search process: a. Initialize whale population, each individual represents a set of threshold combination (such as early warning area threshold 0.85, alert area 0.55, and repelling area 0.45).

[0035] b. Adjust the search step through the nonlinear convergence factor, update the position combined with the sinusoidal spiral strategy, and iteratively find the optimal threshold combination.

[0036] c. Introduce elite reservation strategy to reserve the best individual of each generation, avoiding local optimal trap.

[0037] Verification and dynamic adjustment: a. Verify the biological authenticity of threshold setting through field experiments or controlled experiments, for example, after setting the threshold value of the repelling area to 0.45, observe whether the actual biological dispersal effect meets the expectation.

[0038] b. Dynamic update of threshold value combined with fuzzy logic algorithm: when the biological activity pattern changes (such as seasonal migration), adjust the threshold range through fuzzy rules (such as "if the biological density increases by 20%, the threshold value decreases by 0.1").

[0039] When real-time biological data (such as density, activity frequency) exceeds the current threshold, adjust the threshold value through fuzzy rules. For example: Input: biological density deviation , activity frequency deviation ; Output: Threshold adjustment amount ; Rule: If and , then c. The integrated LSTM-HMM hybrid model predicts biological activity trends and adjusts the threshold in advance to adapt to future changes.

[0040] Preferably, TIWOA has better population diversity in multimodal function optimization, which can avoid local optimal trap; at the same time, the mathematical model fusion experiment verification (such as field investigation + control experiment) in ecological safety threshold research can improve the biological authenticity of threshold setting, and ensure that the threshold distribution of warning area, alert area, displacement area and observation area is accurately matched with the regional ecological characteristics; Step two, distributed biological sensor network layout and data collection.

[0041] Distributed biological sensors are laid out according to the preset node distribution rule, real-time biological data of each area is collected, including biological species, quantity and activity trajectory, and dynamic error correction is performed on the main sensor data.

[0042] Laying out distributed biological sensors includes laying out main sensors along the boundary of the area according to the Legendre polynomial node distribution rule, the node spacing decreases nonlinearly with the importance of the area, and redundant sensors are added in key areas, and the node spacing is represented as: ; Wherein, is the spacing of the nth node, L is the length of the area boundary, N is the total number of nodes, n is the current node number, and k is the nonlinear decrease index.

[0043] The multiplicative square root extended Kalman filter (MSF-EKF) is used instead of the traditional Kalman filter, which improves the numerical stability through covariance square root decomposition and multiplicative noise processing, Covariance square root decomposition represents the state covariance matrix P as a square root factor S (satisfying P=SST) to avoid overflow problems caused by loss of positive definiteness of the covariance matrix in numerical calculation.

[0044] The redundant sensor data is preferentially used to correct the measurement deviation of the main sensor; at the same time, the EMI static noise circuit (composed of ferrite beads and π-type filter) is integrated to reduce the noise of power supply and signal line.

[0045] The core mechanism of MSF-EKF data correction is: Avoid numerical overflow by square root decomposition of state covariance, Introduce a multiplicative noise model to handle nonlinear errors of sensors, and the correction formula is a coupled process of state update and covariance iteration, that is, based on the state estimation at the previous moment and the current redundant sensor measurement value, High-precision biological state estimation is achieved by recursive implementation of square root filtering.

[0046] The collected biological data is subjected to multi-scale noise reduction processing, the biological safety threshold of each region is dynamically updated based on fuzzy logic algorithm, and the control instruction is generated.

[0047] Step three, multi-scale noise reduction of biological data and dynamic threshold optimization.

[0048] For multi-sensor data in the same region, variational mode decomposition (VMD) is used instead of traditional wavelet transform. The signal is adaptively decomposed into multiple intrinsic mode functions with specific bandwidth through a non-recursive variational framework, avoiding the subjectivity of wavelet basis selection. Combined with empirical mode decomposition (EMD), the biological signal is subjected to secondary processing, and the adaptive decomposition characteristic has better feature capturing ability for biological data which is a kind of nonlinear and non-stationary signal. Then, an autoencoder (AE) is introduced for feature compression and reconstruction. The signal features after decomposition are mapped to a low-dimensional space through the encoder to separate noise, and then reconstructed by the decoder to obtain high-precision noise-reduced biological data. The reconstructed data is input into the dynamic threshold calculation unit, and the fuzzy logic algorithm and biological activity time series prediction results are combined to dynamically update the biological safety threshold range of each region. The risk deviation signal is generated by difference calculation of real-time biological data and dynamic threshold value in each region. Based on historical data, when the risk difference is greater than or equal to the risk difference threshold, an emergency dispersal instruction is generated, and when the risk difference is less than the risk difference threshold, a gradual warning instruction is generated. The core logic of autoencoder reconstruction noise reduction data is to train the network with the objective of minimizing reconstruction error, and the encoder weight matrix is optimized through gradient descent iteration to ensure that the key features of biological activity (species, quantity, trajectory) are retained while noise is suppressed. LSTM-HMM hybrid model and improved spatio-temporal convolutional neural network are integrated for training and application: the training process includes inputting multi-region historical biological activity sequence and control record to construct training dataset, first capturing the long-term dependence of biological activity time series (such as population migration rhythm, invasion time series features) through LSTM layer, then inputting the high-dimensional time series features output by LSTM into the spatio-temporal convolution module with fusion residual connection and CBAM attention mechanism—through residual connection to alleviate the gradient vanishing problem of deep network, and enhance the feature propagation ability; embedding CBAM attention mechanism optimizes the channel dimension (highlighting biological species feature weight) and spatial dimension (strengthening high-density activity region feature) in turn, which significantly improves the key feature extraction accuracy. The model training stage adopts a cyclic learning rate scheduling strategy, with the learning rate periodically and dynamically adjusted in the interval [0.001, 0.01], and the label smoothing technique (with a smoothing coefficient of 0.1) is used to soften the hard labels, effectively alleviating the overfitting problem; to adapt to resource-constrained scenarios (such as edge computing nodes), a knowledge distillation mechanism is introduced: the trained deep spatio-temporal convolutional network is used as the teacher model, and by minimizing the output distribution difference (KL divergence loss) between the student model (lightweight network) and the teacher model, knowledge transfer is achieved, with the model parameters compressed by more than 60% while the accuracy loss is less than 3%; the biological safety threshold and control response parameters are dynamically optimized based on the prediction results of the optimized model, improving the adaptability of the threshold to biological activity changes; the loss optimization is expressed as: ; wherein, is the comprehensive loss function (including data fitting loss, knowledge distillation loss, and regularization term), is the KL divergence loss of the teacher model and the student model, is the total variation regularization term of the weight matrix W.

[0049] Step four, biological invasion risk prediction and ecological carrying capacity verification.

[0050] The LSTM-HMM hybrid model is used to analyze the biological invasion risk, and the ecological carrying capacity model is used to verify the compliance of biological distribution, triggering an early warning signal; the LSTM-HMM hybrid model replaces the traditional HMM, and the LSTM is used to extract the dynamic evolution characteristics of biological activity based on the modeling advantage of long-time series data, and then input into the HMM to calculate the current biological invasion probability, wherein the state transition probability of HMM is generated by joint training of LSTM output time series features and historical biological activity data, which can effectively improve the time series accuracy of probability prediction; the Viterbi algorithm is introduced to optimize the state sequence prediction process of HMM, and the optimal state path is found through dynamic programming to avoid the problem of invalid path caused by zero state transition probability; the biological distribution is verified whether it meets the constraints of the ecological carrying capacity model, the PINN (Physics-Informed Neural Network) is used to replace the traditional finite element method to solve the biological diffusion equation, and the adaptive grid refinement technology is used to realize efficient and accurate calculation of biological distribution density in the region, and when either the invasion probability or the ecological carrying capacity violation meets the condition, the early warning signal is triggered and the control instruction is generated; The formula for calculating the current biological invasion probability is: ; wherein, M is the number of states, is the Gaussian mixture weight, is the mean matrix fused with LSTM time series features, is the covariance matrix, is the biological density value at time r. Ecological carrying capacity verification includes efficient solving by PINN fusion biological diffusion physical equation: PINN takes multilayer neural network as carrier, embeds biological diffusion equation as physical constraint into loss function, optimizes data fitting error and physical equation residual through back propagation at the same time, realizes continuous prediction of biological density space-time distribution; combined with adaptive grid encryption technology, automatically adjusts the density of calculation nodes according to the biological density gradient, encrypts the grid in the local high gradient area (such as biological aggregation boundary), so as to improve the calculation precision; the PINN solving constraint of biological diffusion equation is: Physical equation: ; PINN loss term: ; Wherein, D is the diffusion coefficient; R(Z) is the reproduction / extinction term; Z is the PINN predicted biological density; (Z is the physical residual, which needs to be minimized in training.

[0051] When the local biological density predicted by PINN exceeds the set threshold, it is determined that the local biological density is abnormal, and when it is less than or equal to the set threshold, it is determined that the local biological density is normal; according to the regulation instruction, adjust the power of the sound and light equipment in the over-limit area, and synchronously adjust the working mode of the sound and light equipment in different areas through linkage control; Step five, intelligent sound and light system regulation and linkage.

[0052] Based on the risk deviation signal, PID neural network (PIDNN) and sliding mode control fusion algorithm are adopted, and model predictive control (MPC) is introduced to optimize the sound and light power adjustment and regional linkage effect: PIDNN replaces the traditional PID parameter adjustment mechanism, its input layer receives the deviation of real-time biological data and threshold value, error change rate, the hidden layer processes the characteristics through sigmoid activation function, and the output layer directly outputs the dynamically optimized proportional gain, integral gain and differential gain, realizing online adaptive adjustment of PID parameters of neural network; the control quantity output by PIDNN is used as the initial input of sliding mode control, combined with the strong robustness of sliding mode control to suppress system disturbance, through the design of continuous switching function to replace the traditional sign function, the chattering phenomenon in power adjustment process is significantly reduced; for the multi-constrained scene of sound and light power dynamic adjustment, MPC is introduced to handle the constraints of power upper limit, regional ecological characteristics adaptation and multi-region linkage timing, etc., through rolling optimization to predict the power change sequence of future 5-10 time steps, to ensure the coherence of current adjustment action and subsequent response; The PIDNN parameter optimization logic is: taking the minimization of dispersion effect deviation as the goal, the network weight is updated through back propagation algorithm, and the weight update rate is set to 0.01-0.05, to ensure the real-time and stability of parameter adjustment; the power adjustment quantity formula optimized by sliding mode control is: ; wherein, ; wherein, is a gain coefficient; is a sliding mode surface of fusion PIDNN control quantity; is a continuous switching function (value range [-1, 1]).

[0053] MPC constraint processing includes: setting the maximum output power of the sound-light device as a hard constraint, and the power threshold corresponding to the regional ecological sensitivity as a soft constraint, and converting the soft constraint into part of the optimization target through a penalty function; the linkage control sound-light device working mode is represented as: ; wherein, is an adjusted working mode parameter; is an initial mode parameter; is a power adjustment quantity optimized by MPC; is a mode adaptation coefficient (0.8 for the warning area and 1.0 for the driving area).

[0054] The beneficial effects of the embodiment are: through the TIWOA optimization threshold adaptation ecological characteristics, combined with distributed sensors, VMD+EMD noise reduction and LSTM-HMM prediction, precise risk assessment and ecological carrying capacity verification are realized, supplemented by PIDNN-sliding mode-MPC fusion control, the scientific nature, response speed and ecological compatibility of the sound-light dispersion are improved, and the efficient and stable operation of the system is ensured.

[0055] The formulas of the present application are dimensionless, and the preset parameters in the formulas are set by the person skilled in the art according to the actual situation.

[0056] The weight coefficient of the present application is used to measure the influence degree of different factors or variables on a certain result or decision. The definition of weight coefficient is to assign a numerical value to each factor when comparing and evaluating multiple factors to reflect its importance or priority. These weight coefficients can be determined according to specific circumstances and needs, and are usually formulated and confirmed by professionals or relevant parties. By reasonably setting the weight coefficient, the program or system can make more accurate decisions or predictions.

[0057] The above is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art within the technical range disclosed by the present application, according to the technical scheme and the inventive concept of the present application, equivalent replacement or change, should be covered within the protection scope of the present application.

[0058] The preferred embodiments of the application disclosed above are only to facilitate the elucidation of the application. The preferred embodiments do not describe all the details of the application and limit the application to the specific embodiments. Obviously, many modifications and variations can be made in light of the teachings above. The description is chosen and described in order to provide the best illustration of the application and its practical application to those skilled in the art and to enable those skilled in the art to best utilize the application. The application is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent acoustic-optical combined dispersal system based on biosafety threshold regulation, characterized in that, It includes a regional threshold setting module, a sensor data acquisition module, a data processing optimization module, a risk verification module, and an audio-visual control module. Regional threshold setting module: The monitoring area is divided into early warning zone, warning zone, expulsion zone and observation zone. Based on the ecological sensitivity index (ESI) quantification results, the improved whale optimization algorithm TIWOA is introduced. The nonlinear distribution of biosafety threshold is generated through nonlinear convergence factor and sinusoidal spiral update strategy. The threshold is dynamically adjusted by combining fuzzy logic algorithm and LSTM-HMM hybrid model. Sensor data acquisition module: The main sensors are deployed along the region boundary according to the Legendre polynomial node distribution law, and redundant sensors are added in key areas. The node spacing decreases nonlinearly with the importance of the region to correct data errors. An EMI noise reduction circuit containing ferrite beads and π-type filters is integrated to suppress noise. Data processing optimization module: The biological data is decomposed twice using variational mode decomposition (VMD) and empirical mode decomposition (EMD), reconstructed and denoised by autoencoder (AE), and features are extracted by spatiotemporal convolutional network to generate risk bias signals and control instructions. Risk verification module: Calculates the probability of biological invasion using an LSTM-HMM hybrid model, solves the biological diffusion equation, and verifies the compliance of ecological carrying capacity using adaptive grid encryption technology. An early warning is triggered if any condition is exceeded. Acoustic and visual control module: Based on the risk deviation signal, adjust the power and working mode of the acoustic and visual units in each area and achieve linkage.

2. The intelligent acoustic-optical combined dispersal system based on biosafety threshold regulation according to claim 1, characterized in that, The parametric design of TIWOA includes: The nonlinear convergence factor adopts an exponential decay form, with the initial value being the conventional initial value of the whale optimization algorithm. The decay coefficient is determined according to the complexity of the regional ecological sensitivity, with a higher value for highly sensitive mixed regions than for uniform regions. The spiral angle of the sinusoidal spiral strategy changes with the iteration cycle. The amplitude, variation cycle and spiral step size are all adapted to the nonlinear threshold optimization requirements. When the population gets stuck in a local optimum, the amplitude is increased and the step size is decreased in the later stage of the iteration to optimize the threshold. The fitness function balances the threshold-ESI matching degree, threshold nonlinearity, and system response efficiency. The weights are dynamically adjusted according to regional sensitivity. A penalty function with a penalty coefficient is incorporated, and the penalty coefficient increases with the ESI level to strengthen the constraint of sensitive regions.

3. The intelligent acoustic-optical combined dispersal system based on biosafety threshold regulation according to claim 1, characterized in that, In the sensor data acquisition module: MSF-EKF avoids numerical overflow through covariance square root decomposition, introduces a multiplicative noise model to handle sensor nonlinearity errors, and calculates redundant sensor weights based on normalized sensor performance indicators, including historical measurement error, signal-to-noise ratio, and data freshness. The EMI noise reduction circuit uses ferrite beads whose impedance is matched to the noise frequency band of the sensor. The π-type filter contains a combination structure of electrolytic capacitors, ceramic capacitors and inductors, which work together to reduce power supply and signal line noise.

4. The intelligent acoustic-optical combined dispersal system based on biosafety threshold regulation according to claim 1, characterized in that, In the data processing optimization module: VMD adaptively decomposes the signal into multiple intrinsic mode functions, EMD performs secondary processing on nonlinear and non-stationary biological signals, and AE trains with the goal of minimizing reconstruction error, while retaining key features of biological species, quantity and trajectory. The spatiotemporal convolutional network first uses LSTM to capture the temporal dependencies of biological activities, and then uses residual connections to alleviate the gradient vanishing problem in deep networks. The CBAM attention mechanism then sequentially weights and optimizes channel and spatial dimension features to improve the accuracy of key feature extraction.

5. The intelligent acoustic-optical combined dispersal system based on biosafety threshold regulation according to claim 4, characterized in that, The training optimization of the spatiotemporal convolutional network includes: A cyclic learning rate interval adapted to time-series data training is adopted, along with label smoothing techniques to effectively alleviate overfitting; A knowledge distillation mechanism is introduced, using a deep network as the teacher model. Knowledge is transferred to a lightweight student model through KL divergence loss. The comprehensive loss function includes data fitting loss, knowledge distillation loss, and weight matrix regularization term, ensuring that model parameter compression is achieved to adapt to resource-constrained scenarios while meeting accuracy requirements.

6. The intelligent acoustic-optical combined dispersal system based on biosafety threshold regulation according to claim 1, characterized in that, In the risk verification module: The LSTM-HMM hybrid model extracts dynamic evolutionary features of biological activities through LSTM, trains the state transition probabilities of HMM using historical data, and uses the Viterbi algorithm to find the optimal state path through dynamic programming to avoid the problem of invalid paths. The invasion probability is calculated based on the number of states, Gaussian mixture weights, mean matrix, and biological density value.

7. The intelligent acoustic-optical combined dispersal system based on biosafety threshold regulation according to claim 1, characterized in that, In the ecological carrying capacity verification: PINN uses a multi-layer neural network as a carrier and embeds the biological diffusion equation as a physical constraint into the loss function, while optimizing the data fitting error and the physical equation residual. By combining adaptive mesh densification technology, the density of computing nodes is dynamically adjusted according to the biological density gradient, and the mesh is densified in high gradient regions to improve computing accuracy.

8. The intelligent acoustic-optical combined dispersal system based on biosafety threshold regulation according to claim 1, characterized in that, The fusion algorithm of the acoustic-optical modulation module includes: The deviation and error rate of the PIDNN input data from the threshold are processed by the sigmoid activation function to output dynamic PID gain, and the weight update rate is adapted to the real-time and stability requirements of parameter adjustment. Sliding mode control uses the PIDNN output as the initial input and suppresses power adjustment chattering by continuously switching functions; MPC processes power limits, ecological adaptation, and linkage timing constraints to predict time-step power sequences that adapt to dynamic control needs. Linkage mode parameters are calculated based on the initial mode, power adjustment amount, and mode coefficients adapted to regional risk levels.

9. The intelligent acoustic-optical combined dispersal system based on biosafety threshold regulation according to claim 1, characterized in that, The verification and adjustment of the region threshold setting module includes: Verify the biological authenticity of the threshold through field experiments or controlled experiments; When the biological density deviation and the activity frequency deviation change, the threshold adjustment amount is output through fuzzy rules; Integrating LSTM-HMM hybrid models to predict trends in biological activity allows for advance optimization of thresholds to adapt to future changes.

Images