Intelligent Environment Sensing Method and System Based on Outdoor Waterproof TV

Abstract

The present invention relates to an intelligent environment sensing method and system based on an outdoor waterproof TV. The method includes: collecting an environmental monitoring data set of the outdoor waterproof TV; performing non-linear transformation and multi-dimensional feature extraction, and obtaining a fusion feature matrix through feature fusion; calculating weight coefficients of temperature parameters, humidity parameters, light intensity parameters, and rainfall parameters for the fusion feature matrix, and performing weighted calculation with the fusion feature matrix to obtain an environmental risk factor; inputting the environmental risk factor and the fusion feature matrix into a deep neural network based on Proximal Policy Optimization (PPO) for policy iteration optimization calculation to generate a protection control instruction set; respectively generating a driving signal for a display module and a control signal for a waterproof actuator, and executing the driving signal and the control signal through an execution unit. The present invention optimizes the display effect while ensuring the safety of the device through the coordinated control of display parameters and waterproof levels, improving the user viewing experience.

Description

Technical Field

[0001] The present invention relates to the technical field of television environment sensing, and particularly relates to an intelligent environment sensing method and system based on an outdoor waterproof television. Background Art

[0002] With the wide application of outdoor electronic display devices, outdoor waterproof televisions play an important role in public places, advertising displays and other fields. However, the outdoor environmental conditions are complex and changeable, and factors such as temperature, humidity, light intensity and rainfall will have a significant impact on the normal operation and display effect of television devices.

[0003] Traditional outdoor waterproof televisions generally adopt a fixed-threshold protection strategy and a single display parameter adjustment scheme, and cannot make accurate responses to different environmental conditions. This static protection mechanism has a lag in response to sudden weather changes, is prone to insufficient or excessive protection of the device, and affects the service life of the device and the viewing experience. Most of the current environmental sensing solutions on the market rely on simple single-parameter monitoring and preset rule judgment, lacking the ability to comprehensively analyze multiple environmental factors. In practical applications, there are complex interaction relationships between environmental parameters, and an intelligent environmental perception mechanism needs to be established to realize the dynamic weight optimization of environmental parameters and the adaptive adjustment of protection strategies. Summary of the Invention

[0004] The main object of the present invention is to provide an intelligent environment sensing method and system based on an outdoor waterproof television. Through the coordinated control of display parameters and waterproof levels, the present invention optimizes the display effect while ensuring the safety of the device and improves the user viewing experience.

[0005] To achieve the above object, the present invention provides an intelligent environment sensing method based on an outdoor waterproof television, including the following steps:

[0006] Collect an environmental monitoring data set of the outdoor waterproof television through an environmental sensor array;

[0007] Input the environmental monitoring data set into an environmental sensing network for non-linear transformation and multi-dimensional feature extraction, and obtain a fusion feature matrix through feature fusion;

[0008] Calculate the temperature parameter weight coefficient, humidity parameter weight coefficient, light intensity parameter weight coefficient and rainfall parameter weight coefficient for the fusion feature matrix through a Bayesian optimization algorithm, and perform weighted calculation with the fusion feature matrix to obtain an environmental risk factor;

[0009] Input the environmental risk factors and the fusion feature matrix into a deep neural network based on Proximal Policy Optimization (PPO) for policy iteration optimization calculation to generate a set of protection control instructions, where the set of protection control instructions includes display parameter adjustment instructions and waterproof level control instructions;

[0010] Generate a driving signal for the display module and a control signal for the waterproof actuator respectively according to the set of protection control instructions, and execute the driving signal and the control signal through an execution unit.

[0011] The present invention also provides an intelligent environment sensing system based on an outdoor waterproof TV, including:

[0012] An acquisition module, configured to acquire an environmental monitoring data set of the outdoor waterproof TV through an environmental sensor array;

[0013] A feature extraction module, configured to input the environmental monitoring data set into an environmental sensing network for non-linear transformation and multi-dimensional feature extraction, and obtain a fusion feature matrix through feature fusion;

[0014] A weighted calculation module, configured to calculate a temperature parameter weight coefficient, a humidity parameter weight coefficient, a light intensity parameter weight coefficient, and a rainfall parameter weight coefficient for the fusion feature matrix through a Bayesian optimization algorithm, and perform weighted calculation with the fusion feature matrix to obtain environmental risk factors;

[0015] A generation module, configured to input the environmental risk factors and the fusion feature matrix into a deep neural network based on Proximal Policy Optimization (PPO) for policy iteration optimization calculation to generate a set of protection control instructions, where the set of protection control instructions includes display parameter adjustment instructions and waterproof level control instructions;

[0016] An execution module, configured to generate a driving signal for the display module and a control signal for the waterproof actuator respectively according to the set of protection control instructions, and execute the driving signal and the control signal through an execution unit.

[0017] The present invention also provides a computer device, including a memory and a processor, where a computer program is stored in the memory, and when the processor executes the computer program, the steps of the method described in any one of the above are implemented.

[0018] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in any one of the above are implemented.

[0019] In summary, the technical solution provided by the present invention realizes the real-time acquisition and feature extraction of multi-dimensional environmental parameters through the cooperation of the environmental sensor array and the environmental sensing network, improving the accuracy and comprehensiveness of environmental monitoring; uses the Bayesian optimization algorithm to calculate the dynamic weights of environmental parameters, solving the problem that the traditional fixed-weight scheme cannot adapt to environmental changes; based on the strategy optimization mechanism of the PPO deep neural network, realizes the intelligent generation of protection control instructions, significantly reducing the system response delay; through the coordinated control of the display parameters and the waterproof level, optimizes the display effect while ensuring the safety of the device, enhancing the user viewing experience; the multi-layer feature extraction structure of the environmental sensing network enhances the system's perception ability and feature expression ability for complex environmental changes; uses the self-attention mechanism for feature fusion, improving the system's recognition and response ability to key environmental factors; the policy gradient calculation method based on importance sampling ratio clipping ensures the stability and convergence of the control policy update. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a schematic diagram of the steps of an intelligent environmental sensing method based on an outdoor waterproof TV according to an embodiment of the present invention;

[0021] Figure 2 is a front view of an outdoor waterproof TV according to an embodiment of the present invention;

[0022] Figure 3 is a rear view of an outdoor waterproof TV according to an embodiment of the present invention;

[0023] Figure 4 is a structural block diagram of an intelligent environmental sensing system based on an outdoor waterproof TV according to an embodiment of the present invention;

[0024] Figure 5 is a structural schematic diagram of a computer device according to an embodiment of the present invention.

[0025] The realization, functional features and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0027] Referring to Figure 1 , this embodiment provides an intelligent environmental sensing method based on an outdoor waterproof TV, including the following steps:

[0028] S1, collecting an environmental monitoring data set of the outdoor waterproof TV through an environmental sensor array;

[0029] Among them, a preset sampling voltage is applied to the input end of the temperature sensor in the environmental sensor array. This voltage can ensure the stable operation of the temperature sensor while capturing the real temperature change signal in the environment. The output of the temperature sensor is processed by a signal conditioning circuit, which includes steps of amplifying and filtering the analog signal. The design of the amplifier circuit needs to consider the sensitivity of the sensor and the range of the output signal to ensure that small signals can be accurately amplified to a voltage range suitable for subsequent processing, while the filter circuit is used to remove environmental noise and other interference signals to obtain a stable temperature signal. Through this step, the temperature signal after analog-to-digital conversion is converted into a corresponding temperature digital signal. At the same time, an excitation signal is input to the detection end of the humidity sensor in the environmental sensor array. The excitation signal is a carefully designed sine wave or pulse signal, and its frequency and amplitude need to match the optimal working parameters of the sensor. After the signal passes through the humidity sensor, it will respond to the change of environmental humidity. On this basis, impedance matching processing is carried out through a signal conversion circuit to reduce the energy loss in signal transmission and improve the signal quality. During the signal stabilization process, the designed filter and buffer circuit will further eliminate environmental interference so that the output signal can more accurately reflect the actual humidity change. The processed humidity signal is converted into a corresponding humidity digital signal. For the light intensity sensor in the environmental sensor array, using the characteristics of its photosensitive element, under the condition of applying a reverse bias voltage, the conversion of the optical signal to the electrical signal is realized through a photoelectric conversion circuit. When the photosensitive element is illuminated, a current signal proportional to the light intensity is generated. This signal is relatively weak and is amplified through a current amplifier circuit to enhance the signal intensity. In order to make the light intensity signal output by the sensor have better linear characteristics, the signal is corrected through a linearization circuit to ensure that the output light intensity signal can accurately reflect the light change in the environment. These amplified and linearized light intensity signals are converted into light intensity digital signals. During the signal processing of the rain sensor, an AC excitation signal is applied to its detection electrode. The selection of this signal should consider the working characteristics of the rain sensor and environmental conditions, and a high-frequency and low-amplitude AC signal is used to improve the sensitivity. After the sensor receives the external rain information and generates a corresponding response signal, the signal is waveform-shaped through a signal processing circuit to correct the amplitude and frequency drift problems of the rain signal and ensure the standardization and readability of the signal. The processed rain signal is converted into a rain digital signal through analog-to-digital conversion, indicating the real-time change of the rainfall intensity. The temperature digital signal, humidity digital signal, light intensity digital signal, and rain digital signal are collected into the data acquisition module. This module realizes signal switching and sample-and-hold processing through a multiplexer. The design of the multiplexer can effectively coordinate the input of various sensor signals to ensure that each signal can be accurately collected during multi-channel sampling, while the sample-and-hold circuit improves the accuracy of subsequent digital conversion by maintaining the voltage value stability of the signal.By performing digital conversion on multi-channel sampled data, an analog signal is converted into a digital signal using a high-precision analog-to-digital converter. At the same time, data compensation is carried out in combination with an algorithm to correct measurement errors caused by sensor characteristic differences or environmental interference, generating a complete and accurate environmental monitoring data set.

[0030] In this embodiment, Figure 2 It is a front schematic diagram of an outdoor waterproof TV, including: a front shell 1 and a liquid crystal glass 2. The gluing area is in the middle of the front shell 1 and the liquid crystal glass 2. Figure 3 It is a back schematic diagram of an outdoor waterproof TV, including a rear shell 3 and a rear backpack 4. After the rear shell 3 and the rear backpack 4 are assembled, there is a gluing gap in the middle, and the gap is 1 mm, and glass glue is applied.

[0031] S2. Input the environmental monitoring data set into the environmental sensing network for non-linear transformation and multi-dimensional feature extraction, and obtain a fusion feature matrix through feature fusion;

[0032] Specifically, the environmental monitoring data set is input into the input layer of the environmental sensing network, where the data is standardized and time-aligned. Standardization maps environmental data with different dimensions and ranges, such as temperature, humidity, light intensity, and rainfall data, to a unified numerical range (such as [0,1] or [-1,1]) to eliminate the influence caused by dimensional differences and improve the model's adaptability to data. At the same time, the time-alignment method is used to solve the problem of time dimension misalignment caused by different sampling frequencies or data delays during the sensor data acquisition process to ensure the temporal consistency of multi-dimensional data and generate a standardized data sequence with a unified scale and time alignment. The standardized data sequence is input into the reservoir layer of the environmental sensing network for dynamic feature mapping. The reservoir layer consists of 128 neuron nodes, and each node uses the hyperbolic tangent function as the activation function. The hyperbolic tangent function has good non-linear mapping ability and can effectively capture complex non-linear relationships in the input data. At the same time, the input weight matrix of the reservoir layer adopts a sparse connection structure to reduce the parameter scale of the network, lower the computational complexity, and improve the generalization performance of the reservoir. Under the action of the reservoir layer, the standardized data sequence is mapped into a high-dimensional space to generate a reservoir state matrix. The reservoir state matrix can retain the dynamic features of the input data and reflect the potential correlations between data through the internal state evolution of the reservoir. The reservoir state matrix is input into the time feature extraction unit to mine the dynamic information in the time dimension through a recurrent neural network structure. In this process, the recurrent neural network can capture the characteristic patterns of data changing over time through its recurrent connection mechanism, such as the long-term trend of temperature change, the short-term fluctuation characteristics of humidity, the periodic change pattern of light intensity, and the cumulative change law of rainfall. These features are gradually encoded into a set of time feature vectors through recurrent calculations. The time feature vectors not only contain the independent feature information of various environmental data but also integrate the time correlations between them. The time series feature vectors are input into the output layer of the environmental sensing network for feature mapping to obtain a set of environmental feature vectors. The environmental feature vectors are input into the feature fusion network for feature enhancement. In the feature fusion network, non-linear interactions between features are realized through a multi-layer perceptron (MLP) or a convolutional neural network (CNN) structure to improve the expression ability and discriminative ability of the features and generate an enhanced feature matrix. Self-attention weighted combination is performed on the enhanced feature matrix to generate an attention feature matrix. The self-attention mechanism is used to evaluate the importance of each dimension of the enhanced feature matrix and calculate the importance weights of different features for the overall environmental description. The self-attention mechanism can dynamically adjust the weight allocation of features, enabling the model to pay more attention to the feature information most important for the current task. By summing the weighted features, an attention feature matrix is generated. The attention feature matrix is dimension-transformed to generate a fusion feature matrix.The process of dimensionality transformation includes dimensionality reduction or rearrangement of feature dimensions, so that the final fused feature matrix can not only maintain the integrity of the original feature information, but also adapt to the specific requirements of subsequent application scenarios.

[0033] S3. Calculate the weight coefficients of temperature parameters, humidity parameters, light intensity parameters and rainfall parameters for the fused feature matrix through the Bayesian optimization algorithm, and perform weighted calculation with the fused feature matrix to obtain the environmental risk factor.

[0034] It should be noted that the objective function of the Bayesian optimization algorithm is established based on the fused feature matrix. By optimizing the weight distribution of each parameter, the finally calculated risk factor can accurately reflect the comprehensive impact of different environmental conditions on the operation of outdoor waterproof TVs. After constructing the objective function, Gaussian process regression is used to model the weight space of temperature, humidity, light intensity and rainfall parameters. Gaussian process regression constructs a joint distribution model of these parameters based on existing data, and describes the potential correlation and uncertainty between parameters in the weight space through dynamic adjustment of the mean and covariance, thus providing data-driven probabilistic inference support for weight optimization. The four-dimensional joint distribution model generated by the Gaussian process is input into the Bayesian optimization algorithm, and sampling points are selected according to the principle of maximizing the expected improvement. This principle balances the contradiction between exploring new weight combinations and using existing weight information, ensuring that the optimization process can cover the entire parameter space and focus on the possible optimal regions. In each iteration, the Bayesian optimization algorithm predicts the parameter combination that brings the maximum optimization benefit based on the current distribution model and evaluates the sampling. The sampling results are used to update the posterior model of the weight distribution. The iterative method can ensure that the weight optimization process converges to the global optimal solution. As the sampling progresses, the generated weight parameter sample set is used to calculate the sample mean and confidence interval of each parameter. The sample mean reflects the overall trend of the weight parameters, while the confidence interval quantifies the uncertainty of different parameters during the optimization process. Based on the sample mean and confidence interval, the weight parameters are updated using the maximum posterior probability criterion to obtain the optimized weight coefficients of temperature parameters, humidity parameters, light intensity parameters and rainfall parameters. The optimized weights are used to perform weighted combination of each dimension feature in the fused feature matrix, and the importance of different features is integrated into a comprehensive score vector according to the optimized weight ratio. This vector expresses the comprehensive risk impact of various environmental features in a quantitative form. In order to convert the comprehensive score vector into a practically usable environmental risk factor, a piecewise linear mapping processing method is introduced. By setting multiple risk level intervals, the score vector is divided into different levels of risk categories such as low risk, medium risk and high risk. The scores within each risk interval are converted according to a linear ratio, and the finally generated environmental risk factor can effectively reflect the potential impact of the current environmental conditions on outdoor waterproof TVs and provide a clear guiding basis for the protection strategy of the equipment.

[0035] The joint distribution model of four-dimensional parameters is input into the Gaussian kernel function, and the probability density value of each point in the parameter space is calculated through the transformation of the kernel function. The Gaussian kernel function uses its non-linear mapping ability to transform the input weight parameter space into a high-dimensional feature space, thereby capturing the potential complex relationships between parameters. In this process, each data point in the joint distribution model is mapped through the kernel function to generate corresponding probability density values, and these values are arranged in the form of a matrix to form a probability density distribution matrix. This matrix defines the probability weights of each point in the space, reflecting the potential possibilities of different parameter combinations. The parameter space grid is divided for the probability density distribution matrix to generate an initial sampling point set of four-dimensional parameters. The purpose of grid division is to create an initial sampling point framework within the parameter space so that the entire space is evenly covered. The initial sampling point set is input into the expected improvement calculation module to calculate the expected improvement value of each sampling point. The calculation of the expected improvement value comprehensively considers the optimization benefits brought by the current sampling point and the contribution of this point to the exploration of the parameter space, so as to evaluate the importance of this point. After the calculation is completed, a set of expected improvement value sequences is generated, reflecting the priorities of each point in the parameter space. The expected improvement value sequences are sorted in descending order, and the top E points with the highest expected improvement values are selected to form a candidate sampling point set. These candidate sampling point sets contain parameter points that potentially contribute to the target optimization, representing the areas in the current space that are most worthy of further exploration. After the candidate sampling points are input into the local search module, a preset search radius is defined around each candidate point, and the gradient is calculated within this radius range. The goal of gradient calculation is to capture the change trend of the candidate point in the local area, and a local gradient vector is generated through this process. According to the local gradient vector, the candidate sampling points are parameter-updated to generate optimized sampling points. During the update process, the parameters of each candidate point move along the direction indicated by the gradient, and the moving amplitude is jointly determined by the preset learning rate and the gradient magnitude. This update method can ensure that the sampling points gradually approach the optimal solution area while avoiding falling into local extrema. The optimized sampling points are input into the Gaussian process regression model again to update the model parameters and recalculate the posterior distribution. Through this process, the prediction ability of the Gaussian process model is further enhanced, and it can more accurately describe the current parameter space distribution. Iterate the above steps multiple times to converge to a set of weight parameter sample sets.

[0036] S4. Input the environmental risk factors and the fusion feature matrix into the deep neural network based on Proximal Policy Optimization (PPO) for policy iteration optimization calculation to generate a set of protection control instructions, and the set of protection control instructions includes display parameter adjustment instructions and waterproof level control instructions;

[0037] Specifically, the environmental risk factors and the fusion feature matrix are input into the input layer of a deep neural network based on Proximal Policy Optimization (PPO). The input layer is designed as a fully connected layer with 256 neurons, and these neurons perform non-linear transformation through the ReLU activation function. The choice of the ReLU function is to enhance the network's expressive ability and avoid the vanishing gradient problem, generating an initial feature vector. The initial feature vector is input into the shared layer of the deep neural network. The shared layer consists of three fully connected layers, with the number of neurons in each layer being 128, 64, and 32 respectively, and each neuron is also processed using the ReLU activation function. The role of the shared layer is to generate a high-level shared feature representation by gradually extracting feature hierarchies. The design of the shared layer aims to capture the potential associated features and environmental impact patterns in the input data, while ensuring the sharing of key underlying features between different task branches, reducing network redundancy and computational complexity. The shared feature representation is respectively input into the display parameter optimization branch and the waterproof level optimization branch. Each branch contains two fully connected layers, with 16 neurons in each layer, and also uses the ReLU activation function. This branch design enables the network to simultaneously process the features related to display parameter adjustment and the features related to waterproof level control, generating the corresponding display branch feature vector and waterproof branch feature vector respectively. The display branch feature vector and the waterproof branch feature vector are respectively input into the value function network for state value calculation. The main task of the value function network is to evaluate the value distribution in the current state, and through this evaluation, generate the display state value evaluation and the waterproof state value evaluation. The display state value evaluation and the waterproof state value evaluation are input into the policy gradient calculation unit, and through the importance sampling ratio clipping method, which is the core of the PPO algorithm, calculate the display parameter adjustment probability distribution and the protection action probability distribution. The role of the clipping method is to limit the amplitude of policy updates, thereby ensuring the stability and efficiency of policy optimization. These two probability distributions respectively represent the priorities of display parameter adjustment actions and protection actions in the current state. According to the display parameter adjustment probability distribution and the protection action probability distribution, generate the display parameter adjustment strategy and the waterproof level control strategy. These strategies are input into the action mapping module, and through a lookup table, map the discrete actions to continuous control quantities. The design of the lookup table is adjusted according to the control requirements of specific hardware devices, so as to ensure that the generated display parameter adjustment instructions and waterproof level control instructions can accurately match the operation requirements of the actual control devices. Package the display parameter adjustment instructions and the waterproof level control instructions to form a complete protection control instruction set. The instruction set includes specific display parameter adjustment schemes, such as the adjustment of screen brightness, contrast, and display mode, and also includes waterproof level control measures, such as triggering the sealing system, drainage device, or opening the protective cover, etc.

[0038] Based on the display state value evaluation and the waterproof state value evaluation, calculate the ratios of the new strategy to the old strategy respectively. This ratio measures the improvement of the current strategy compared to the previous strategy and reflects the magnitude of the strategy change during the optimization process. By dynamically calculating the optimization weights for display parameter adjustment and waterproof level control, obtain the display parameter adjustment weight and the waterproof level adjustment weight, which reflect the contributions of different strategies to the target optimization in the current state. Compare the display parameter adjustment weight and the waterproof level adjustment weight with the preset clipping range. The clipping range is to limit the magnitude of the strategy update and prevent training instability caused by excessive strategy changes. During this process, use the minimum value function to truncate the display parameter adjustment weight and the waterproof level adjustment weight to obtain the clipped display adjustment weight and the clipped waterproof adjustment weight. The core of clipping is to ensure the stability of the optimization process by balancing exploration and exploitation, while maintaining the adaptability of the strategy in the parameter space. Based on the clipped display adjustment weight and the waterproof adjustment weight, calculate the policy objective function. The design of the policy objective function aims to maximize the reward signal and at the same time constrain the range of policy updates through the clipped weights. During this process, calculate the display parameter policy gradient and the waterproof level policy gradient respectively as the directional guidance for policy optimization. After obtaining the policy gradients, constrain the display parameter policy gradient and the waterproof level policy gradient to limit their change magnitudes and ensure the rationality and feasibility of the update directions, generating the display parameter update direction and the waterproof level update direction. Project the display parameter update direction and the waterproof level update direction into the constraint space through the proximal mapping function. The role of the proximal mapping function is to ensure that the update direction is always within the defined constraint space and avoid parameter instability problems caused by excessive gradient update magnitudes. After this projection process, obtain the display parameter update amount and the waterproof level update amount within the constraint space. According to the display parameter update amount and the waterproof level update amount within the constraint space, update the parameters of the policy network to generate the updated display policy network and the waterproof policy network. The updated policy network has stronger environmental adaptability and optimization capabilities and can handle dynamic changing environmental conditions more efficiently. To calculate the actual control actions, input the updated display policy network and the waterproof policy network into the Actor network for forward calculation. The Actor network is responsible for extracting the display parameter action space and the waterproof level action space from the policy network, and these action spaces define all possible control actions and their corresponding priorities. Perform probability normalization on the display parameter action space and the waterproof level action space to ensure that the sum of the probability distributions of all control actions is 1, obtaining the display parameter adjustment probability distribution and the protection action probability distribution.

[0039] S5. Generate the drive signal for the display module and the control signal for the waterproof actuator respectively according to the protection control instruction set, and execute the drive signal and the control signal through the execution unit.

[0040] Among them, the protection control instruction set is generated by the system's deep neural network and optimization algorithm, including display parameter adjustment instructions and waterproof level control instructions. These instructions have been optimized by intelligent algorithms logically and have dynamic adaptability to the current environment. After receiving the instructions, it enters the instruction parsing stage, and the parsing module maps each instruction in the protection control instruction set to specific hardware operation logics. For example, the display parameter adjustment instructions include multiple adjustment parameters such as screen brightness, contrast, and color temperature, while the waterproof level control instructions involve specific actions such as the opening and closing of the waterproof cover plate, the startup of the drainage device, and the pressurization of the sealing system. The parsing module converts the content of these instructions into specific signal types, signal amplitudes, and the addresses of target devices to ensure that the hardware devices can respond correctly. Input the instruction parsing result into the signal generation module. In the signal generation stage, corresponding control signals are generated for different execution devices according to the parsed parameter content. For the driving signal of the display module, by adjusting the parameter values in the display parameter adjustment instructions, such as the percentage of brightness or the set value of contrast, it is mapped into a PWM (pulse width modulation) signal or a digital level signal. These signals are matched with the driving circuit of the display module, enabling the brightness, color temperature, and display effect of the screen to be dynamically adjusted according to user expectations or environmental requirements. During the signal generation process, filtering and calibration are performed to ensure that the generated driving signals meet the specification requirements of the hardware devices in terms of amplitude and frequency, avoiding overload or invalid operations. For the control signal of the waterproof actuator, the signal generation module generates corresponding control logic signals according to the level requirements in the waterproof level control instructions. These signals include switch-type signals (such as relay control signals) for triggering the opening and closing operations of the waterproof cover plate; analog signals (such as voltage or current signals) for controlling the pressure adjustment of the hydraulic or pneumatic sealing system. For waterproof devices that require continuous adjustment, such as intelligent drainage systems, the control signals are generated in the form of proportional adjustment to precisely control the operating speed or drainage volume of the devices. In this stage, the signals are corrected according to the response characteristics of the devices, such as considering delay compensation or dynamic range limitation, to ensure that the signals can drive the devices quickly and accurately. Input the generated driving signals and control signals into the execution unit. The execution unit converts the signals generated by the system into physical actions that the actual devices can execute. During the execution of the display module, the execution unit converts the PWM signal into a current or voltage driving signal and loads it into the LED backlight system or liquid crystal panel of the display module through a driving chip. At the same time, the status monitoring circuit in the execution unit monitors the response status of the display module in real time to ensure that the result of the parameter adjustment is consistent with the expectation. For the waterproof actuator, the execution unit adapts the control signals according to the device type. For example, when controlling an electric waterproof cover plate, the execution unit converts the digital signal into the switch operation of the device through a relay module and uses a position sensor to feedback the current state to prevent over-operation or jamming phenomena.For a hydraulic sealing system, the execution unit adjusts the operating state of the hydraulic pump according to the input analog signal to keep the pressure within a specified range. The execution unit also has a protection mechanism. When abnormal signals or abnormal equipment operating states are detected, it will immediately trigger the fault protection logic to avoid further damage to the equipment or affecting the normal operation of the system. The entire system ensures the consistency between the display module and the waterproof actuator's response and the protection control instruction set through signal closed-loop control. The status feedback of the execution unit is transmitted back to the system main control unit for dynamically adjusting the generation logic of subsequent signals. For example, when the environmental conditions change drastically, the system can update the protection control instruction set in real time according to the latest environmental sensing data and generate new drive signals and control signals, thus achieving an adaptive dynamic response to the environment.

[0041] In one example, an environmental monitoring data set of an outdoor waterproof TV is collected through an environmental sensor array, including:

[0042] Apply a preset sampling voltage to the input end of the temperature sensor in the environmental sensor array, and perform analog signal amplification and filtering processing through a signal conditioning circuit to obtain a temperature digital signal;

[0043] Input an excitation signal to the detection end of the humidity sensor in the environmental sensor array, and perform impedance matching and signal stabilization processing through a signal conversion circuit to obtain a humidity digital signal;

[0044] Apply a reverse bias voltage to the photosensitive element of the light intensity sensor in the environmental sensor array, and perform current amplification and linearization processing through a photoelectric conversion circuit to obtain a light intensity digital signal;

[0045] Apply an AC excitation signal to the detection electrode of the rain sensor in the environmental sensor array, and perform waveform shaping and digitization processing through a signal processing circuit to obtain a rain digital signal;

[0046] Input the temperature digital signal, humidity digital signal, light intensity digital signal, and rain digital signal into the data acquisition module, and perform signal switching and sample and hold processing through a multiplexer to obtain multi-channel sampled data;

[0047] Perform digital quantity conversion and data compensation on the multi-channel sampled data to obtain the environmental monitoring data set of the outdoor waterproof TV.

[0048] In this example, a stable preset sampling voltage is applied to the input end of the temperature sensor, denoted by the symbol This is the drive voltage for the sensor to work, and its magnitude is selected according to the specifications of the sensor. For example, It is maintained at a fixed value to ensure the stable operation of the sensor. The output signal of the temperature sensor is an analog voltage that has a linear relationship with the environmental temperature , and its expression is ;

[0049] Among them, represents the sensitivity of the sensor, and represents the amplitude of the voltage change caused by the temperature change; is the zero-point offset, that is, when the temperature is , the reference voltage value. This signal is amplified and filtered by the signal conditioning circuit. The gain of the amplifier circuit is represented by , and its function is to boost the weak signal output by the sensor to a range suitable for the input of the analog-to-digital converter (ADC): ;

[0050] The filtering module removes high-frequency noise in the signal by designing an appropriate cut-off frequency. The processed signal is converted into a digital signal by the ADC, and the digital signal The calculation formula is: ;

[0051] Among them, is the reference voltage of the ADC, is the resolution of the ADC (such as when it is 12 bits), and the digital signal is used to represent the numerical result of the current temperature. The humidity sensor needs to input an excitation signal at its detection end, and a sine wave signal is adopted, which is expressed as: ;

[0052] Among them, is the amplitude of the signal, is the frequency of the signal. The output signal of the humidity sensor is linearly related to the environmental relative humidity , and its expression is: ;

[0053] Among them, is the sensitivity of the humidity sensor; is the offset voltage value. The signal is processed by the impedance matching circuit to reduce the signal source impedance to adapt to the subsequent signal processing circuit, and at the same time, the noise and signal fluctuations are eliminated by the stabilization circuit. After analog-to-digital conversion, the humidity digital signal is generated:

[0054] ;

[0055] For the light intensity sensor, its photosensitive element needs to be applied with a reverse bias voltage , and its function is to enhance the response sensitivity of the photosensitive element. The output of the sensor is the photocurrent , and its magnitude is related to the light intensity Is directly proportional, expressed as: ;

[0056] Wherein, Is the optoelectronic response coefficient of the sensor. The photocurrent is converted into a voltage signal by a current amplifier : ;

[0057] Where Is the gain of the amplifier. After linearization processing, the signal is digitized into an illumination intensity digital signal by an analog-to-digital converter : ;

[0058] For the rain sensor, an AC excitation signal needs to be applied to its detection electrode , Expressed as: ;

[0059] Where Is the signal amplitude, Is the signal frequency. The output signal of the rain sensor And the rainfall The relationship is: ;

[0060] Where Is the sensitivity of the rain sensor; Is the reference voltage. After the signal passes through the waveform shaping circuit to remove the non-linear components, it is converted into a rain digital signal through the ADC : ;

[0061] All the generated digital signals , , And Are input into the data acquisition module. Signal switching and sample-and-hold processing are performed through a multiplexer. The multiplexer sequentially acquires the signals of each channel according to a preset time series, while the sample-and-hold circuit ensures that the voltage value of the sampling point remains stable during the switching process. After the multi-channel data collected is corrected for the non-linear deviation and system error of the sensor through a digital compensation algorithm, an environmental monitoring data set is generated: ;

[0062] This data set serves as the core input for environmental analysis and provides accurate environmental information for the intelligent regulation of the device.

[0063] In an example, the environmental monitoring data set is input into the environmental sensing network for non-linear transformation and multi-dimensional feature extraction, and a fused feature matrix is obtained through feature fusion, including:

[0064] Input the environmental monitoring data set into the input layer of the environmental sensing network for data standardization and time series alignment to obtain a standardized data sequence;

[0065] Input the standardized data sequence into the reservoir layer of the environmental sensing network for dynamic feature mapping. The reservoir layer contains 128 neuron nodes, and each neuron node uses the hyperbolic tangent function as the activation function. The input weight matrix adopts a sparse connection structure to obtain a reservoir state matrix;

[0066] Input the reservoir state matrix into the time feature extraction unit, and extract the temperature change trend, humidity fluctuation characteristics, light intensity change pattern and rainfall accumulation characteristics through a recurrent neural network structure to obtain a time series feature vector;

[0067] Input the time series feature vector into the output layer of the environmental sensing network for feature mapping to obtain an environmental feature vector, and input the environmental feature vector into the feature fusion network for feature enhancement to obtain an enhanced feature matrix;

[0068] Perform self-attention weighted combination on the enhanced feature matrix to obtain an attention feature matrix, and perform dimensionality transformation on the attention feature matrix to generate a fusion feature matrix.

[0069] In this example, the environmental monitoring data set contains sampling data of various environmental parameters, such as temperature, humidity, light intensity and rainfall, which are respectively denoted as . These data need to be standardized and time series aligned before being input into the environmental sensing network to eliminate the influence of data dimension differences and sampling delays. Map the value ranges of each variable to [0,1] or a standard normal distribution, which is achieved through the following formula: ;

[0070] where, represents the standardized data, and are the mean and standard deviation of the data respectively. For time series alignment, interpolation or dynamic time warping algorithms are used to align the time series of different sensors to the same time step to form a standardized data sequence . Input the standardized data sequence into the reservoir layer of the environmental sensing network. The reservoir layer contains neuron nodes, and the activation function of each node uses the hyperbolic tangent function (tanh), and its expression is: ;

[0071] The weight matrix input into the reservoir is a sparse connection structure matrix, where the sparsity rate controls the proportion of non-zero elements to improve the calculation efficiency and prevent overfitting. The reservoir state matrix The calculation formula is: ;

[0072] Among them, is the bias vector. The reservoir state matrix extracts the high-dimensional dynamic features of the input data while retaining the temporal correlation. The reservoir state matrix is input into the time feature extraction unit, which uses a recurrent neural network structure to extract temporal features, such as temperature change trends, humidity fluctuation features, light intensity change patterns, and rainfall accumulation features. The recurrent neural network captures the dependencies in the time series through cyclic connections, and its state update formula is: ;

[0073] Among them, is the hidden state vector at time , and are the weight matrices of the input and hidden states respectively, is the activation function, choosing Sigmoid or ReLU. The finally output temporal feature vector represents the key patterns of the time series. The temporal feature vector is input into the output layer of the environmental sensing network, and through feature mapping, an environmental feature vector is generated. The feature mapping process is implemented through a fully connected layer, and the formula is: ;

[0074] Among them, is the activation function, such as ReLU, and are the weight and bias respectively. To enhance the feature expression ability, the environmental feature vector is input into the feature fusion network, and through non-linear combination, an enhanced feature matrix is generated. The fusion network contains multiple hidden layers, and the output of each layer is input into the next layer after being activated by ReLU. Its calculation formula is: ;

[0075] The finally output enhanced feature matrix combines the information of multi-dimensional features. The self-attention mechanism is applied to the enhanced feature matrix for weighted combination. The attention weights are calculated by the following formula: ;

[0076] Among them, and are the query vector and key vector respectively, and the attention weight is used to weight each element of the matrix. The attention feature matrix after weighted combination is: ;

[0077] Among them, is the value vector. Perform dimensional transformation on the attention feature matrix to generate the fused feature matrix by dimensionality reduction or rearrangement. For example, use the principal component analysis method to select the principal feature components, and its formula is: ;

[0078] Among them, is the projection matrix of principal component analysis, and the finally output is the fused feature matrix, which has stronger feature expression ability and discriminability.

[0079] In one example, calculate the temperature parameter weight coefficient, humidity parameter weight coefficient, light intensity parameter weight coefficient, and rainfall parameter weight coefficient for the fused feature matrix through the Bayesian optimization algorithm, and perform weighted calculation with the fused feature matrix to obtain the environmental risk factor, including:

[0080] Construct the objective function of the Bayesian optimization algorithm based on the fused feature matrix, and perform weight space modeling on the temperature parameter, humidity parameter, light intensity parameter, and rainfall parameter through Gaussian process regression to obtain the joint distribution model of the four-dimensional parameters;

[0081] Input the joint distribution model of the four-dimensional parameters into the Bayesian optimization algorithm, select the sampling points based on the principle of maximizing the expected improvement, and perform iterative sampling on the weight parameter space to obtain the weight parameter sample set;

[0082] Calculate the sample mean and confidence interval according to the weight parameter sample set, and update the weight coefficient through the maximum posterior probability criterion to obtain the temperature parameter weight coefficient, humidity parameter weight coefficient, light intensity parameter weight coefficient, and rainfall parameter weight coefficient;

[0083] Based on the temperature parameter weight coefficient, humidity parameter weight coefficient, light intensity parameter weight coefficient, and rainfall parameter weight coefficient, perform multi-parameter weighted combination on the fused feature matrix to obtain the comprehensive score vector;

[0084] Perform hierarchical mapping on the comprehensive score vector through the piecewise linear mapping function to obtain the environmental risk factor.

[0085] In this example, according to the fused feature matrix , construct the objective function of the Bayesian optimization algorithm to optimize the weight allocation of each feature parameter. The fused feature matrix is a multi-dimensional matrix, the rows of which represent samples, and the columns represent feature dimensions, corresponding to temperature , humidity , light intensity and rainfall 。The objective function is defined as the combined effect of parameter weights, used to describe the contribution of weights to environmental risk, expressed as: ;

[0086] where are the weight coefficients of temperature, humidity, light intensity, and rainfall respectively, is the th corresponding eigenvalue of the sample, is the number of samples. To optimize the weight coefficients, the objective function is modeled through Gaussian process regression. The Gaussian process assumes that the weight distribution of each feature follows a joint Gaussian distribution, with a mean of and a covariance matrix of . The joint distribution is expressed as: ;

[0087] where . Through the Gaussian kernel function , the similarity between input data points is defined, and the kernel function form is: ;

[0088] where are data points, is the kernel width parameter. The Gaussian process uses the kernel function to calculate the covariance matrix and updates the posterior distribution based on the observed data. The joint distribution model of the Gaussian process is input into the Bayesian optimization algorithm, and sampling points are selected based on the principle of maximizing the expected improvement. The goal of the expected improvement is to find the point in the weight space that is most likely to improve the objective function value. For each candidate point , its expected improvement value is defined as: ;

[0089] where is the current optimal value, represents the expected value. By optimizing , the next sampling point is selected. During the sampling process, the algorithm iteratively updates the posterior distribution of the weight parameters, generating a sample set of weight parameters through each sampling, where is the number of samplings. The sample mean and the confidence interval are calculated based on the sample set, and the confidence interval is defined as: ;

[0090] where and are the sample mean and standard deviation respectively, is the confidence coefficient. Finally, the weight coefficients Based on the optimized weight coefficients, a multi-parameter weighted combination is performed on the fused feature matrix to generate a comprehensive scoring vector :

[0091] ;

[0092] Among them, , each element represents the comprehensive score of the th sample. The comprehensive scoring vector is mapped to an environmental risk factor through a piecewise linear mapping function. The piecewise linear mapping function is defined as follows:

[0093] ;

[0094] Among them, and are the mapping coefficients of different risk intervals, is the piecewise boundary value. The finally output environmental risk factor is used to evaluate the risk level of the current environment.

[0095] In an example, the joint distribution model of four-dimensional parameters is input into the Bayesian optimization algorithm. Sampling points are selected based on the principle of maximizing expected improvement, and iterative sampling is performed on the weight parameter space to obtain a weight parameter sample set, including:

[0096] The joint distribution model of four-dimensional parameters is input into the Gaussian kernel function, and the probability density values of each point in the parameter space are calculated through kernel function transformation to obtain a probability density distribution matrix;

[0097] The probability density distribution matrix is partitioned into a parameter space grid to generate an initial sampling point set of four-dimensional parameters, and the initial sampling point set is input into the expected improvement calculation module to calculate the expected improvement value of each sampling point to obtain an expected improvement value sequence;

[0098] The expected improvement value sequence is sorted in descending order, and the top E points with the highest expected improvement values are selected as candidate sampling points to obtain a candidate sampling point set, and the candidate sampling point set is input into the local search module to perform gradient calculation around each candidate point with a preset search radius to obtain a local gradient vector;

[0099] The candidate sampling points are updated according to the local gradient vector to obtain optimized sampling points, and the optimized sampling points are input into the Gaussian process regression model to update the model parameters and recalculate the posterior distribution to obtain a weight parameter sample set.

[0100] In this example, the joint distribution model of four-dimensional parameters is input into the Gaussian kernel function for processing to calculate the probability density value of each point in the parameter space. Assume that the four-dimensional parameters are temperature weights , humidity weight , light intensity weight and rainfall weight , which together constitute a joint distribution model . The Gaussian kernel function is defined as: ;

[0101] where and are any two sets of parameters, is the kernel width, which controls the sensitivity of the kernel function to distance. By calculating the kernel function values for all parameter combinations, a probability density distribution matrix K is constructed, and its element represents the similarity between the th and the th parameter points. The parameter space of the probability density distribution matrix is partitioned into a grid to generate an initial sampling point set. The parameter space is discretized into a multi-dimensional grid. Assuming that each dimension is divided into points, the total number of sampling points is . The grid point is expressed as: ;

[0102] The probability density value corresponding to each grid point is obtained through the normalization calculation of the Gaussian kernel function matrix: ;

[0103] The initial sampling point set is input into the expected improvement calculation module to evaluate the optimization potential of each point. For each sampling point , its expected improvement value is defined as: ;

[0104] where is the objective function value, is the current optimal value, represents the expected value. Through Gaussian process prediction of the mean and standard deviation , is expressed as: ;

[0105] where , is the cumulative distribution function of the standard normal distribution, is the probability density function of the standard normal distribution. Calculate the expected improvement values for all initial sampling points, generate an expected improvement value sequence, and sort them in descending order. Select the top These points serve as the candidate sampling point set. Each candidate point represents a potential efficient sampling point. The candidate sampling point set is input into the local search module, and gradient calculations are performed around each candidate point with a preset search radius . The gradient vector represents the local change direction of the objective function at point , and its calculation formula is:

[0106] ;

[0107] Update the positions of the candidate sampling points according to the gradient vector. The updated points are expressed as: ;

[0108] where is the learning rate, which controls the update step size. Input the optimized sampling points into the Gaussian process regression model, update the posterior distribution of the model, and recalculate the mean and covariance matrix of the parameters. By iterating the above process, a sample set of weight parameters is generated, where each sample point represents a set of optimized parameter weights. The distribution of the sample set of weight parameters reflects the optimization result of the parameter space and is used for further analysis or decision-making.

[0109] In one example, the environmental risk factor and the fusion feature matrix are input into a deep neural network based on Proximal Policy Optimization (PPO) for policy iteration optimization calculation to generate a set of protection control instructions. The set of protection control instructions includes display parameter adjustment instructions and waterproof level control instructions, including:

[0110] Input the environmental risk factor and the fusion feature matrix into the input layer of the deep neural network based on Proximal Policy Optimization (PPO). The input layer contains 256 neurons, and through non-linear transformation using the ReLU activation function, an initial feature vector is obtained;

[0111] Input the initial feature vector into the shared layer of the deep neural network based on Proximal Policy Optimization (PPO). The shared layer contains 3 fully connected layers, with each layer containing 128, 64, and 32 neurons respectively. Each neuron uses the ReLU activation function to obtain a shared feature representation;

[0112] Input the shared feature representation into the display parameter optimization branch and the waterproof level optimization branch respectively. The display parameter optimization branch and the waterproof level optimization branch contain 2 fully connected layers, with each fully connected layer containing 16 neurons. Using the ReLU activation function, a display branch feature vector and a waterproof branch feature vector are obtained;

[0113] Input the display branch feature vector and the waterproof branch feature vector into the value function network for state value calculation respectively to obtain a display state value evaluation and a waterproof state value evaluation;

[0114] The display status value evaluation and the waterproof status value evaluation are respectively input into the policy gradient calculation unit, and the policy gradient is calculated by the importance sampling ratio clipping method to obtain the display parameter adjustment probability distribution and the protection action probability distribution;

[0115] Based on the display parameter adjustment probability distribution and the protection action probability distribution, a display parameter adjustment strategy and a waterproof level control strategy are generated, and the display parameter adjustment strategy and the waterproof level control strategy are input into the action mapping module. The discrete actions are mapped into continuous control quantities through a lookup table to obtain a display parameter adjustment instruction and a waterproof level control instruction;

[0116] The display parameter adjustment instruction and the waterproof level control instruction are encapsulated to generate a protection control instruction set.

[0117] In this example, the environmental risk factor and the fused feature matrix are input into the input layer of the deep neural network based on proximal policy optimization. The fused feature matrix contains multi-dimensional features of different environmental parameters (such as temperature, humidity, light intensity, and rainfall), expressed as , where each row is a feature vector. The input layer contains 256 neurons for performing a preliminary non-linear transformation on the input data. The output of each neuron is calculated through the ReLU activation function, and its formula is:

[0118] ;

[0119] where, is the input vector, and are the weight and bias of the th neuron in the input layer respectively, is the weighted input of the neuron. After passing through the input layer, an initial feature vector is output as the input for the subsequent network layers. The initial feature vector is input into the shared layer. The shared layer contains three fully connected layers, with 128, 64, and 32 neurons in each layer respectively. The output of each layer is calculated through the ReLU activation function. The goal of the shared layer is to extract the global feature representation, gradually compress the input high-dimensional feature space while retaining important information. The output formula of the shared layer is:

[0120] ;

[0121] where, and are the weights and biases of the th layer. The shared feature representation is obtained , for task branches. Input the shared feature representation into the display parameter optimization branch and the waterproof level optimization branch respectively. These two branches each contain two fully connected layers, with 16 neurons in each layer, and output the display branch feature vector and the waterproof branch feature vector . Input them into the value function network respectively for state value calculation. The value function network estimates the value of each state , and the formula is:

[0122] ;

[0123] ;

[0124] Obtain the display state value evaluation and the waterproof state value evaluation . Input the state value evaluation into the policy gradient calculation unit, and calculate the policy gradient by combining the importance sampling ratio clipping method. The sampling ratio is defined as: ;

[0125] where, is the probability distribution of the current policy, is the probability distribution of the old policy. Clip the policy gradient to constrain the policy update amplitude and prevent unstable optimization. By optimizing the policy objective function, obtain the display parameter adjustment probability distribution and the protection action probability distribution . Generate the display parameter adjustment policy and the waterproof level control policy based on the probability distribution, and input these policies into the action mapping module. The action mapping module maps the discrete action to the continuous control quantity : ;

[0126] Obtain the display parameter adjustment instruction and the waterproof level control instruction. Package these instructions to form a protection control instruction set. This instruction set is directly used to drive the device and adjust the display parameters (such as brightness, contrast) and waterproof protection measures (such as sealing pressure or drainage device) in real time.

[0127] In one example, input the display state value evaluation and the waterproof state value evaluation into the policy gradient calculation unit respectively, calculate the policy gradient by the importance sampling ratio clipping method, and obtain the display parameter adjustment probability distribution and the protection action probability distribution, including:

[0128] Based on the display state value evaluation and the waterproof state value evaluation, calculate the new and old policy ratios respectively to obtain the display parameter adjustment weight and the waterproof level adjustment weight;

[0129] Compare the display parameter adjustment weight and the waterproof level adjustment weight with the preset cropping range, and perform truncation processing through the minimum value function to obtain the cropped display adjustment weight and the cropped waterproof adjustment weight;

[0130] Calculate the policy objective function based on the cropped display adjustment weight and the cropped waterproof adjustment weight to obtain the display parameter policy gradient and the waterproof level policy gradient, and constrain the display parameter policy gradient and the waterproof level policy gradient to obtain the display parameter update direction and the waterproof level update direction;

[0131] Project the display parameter update direction and the waterproof level update direction through the proximal mapping function to obtain the display parameter update amount within the constraint space and the waterproof level update amount within the constraint space;

[0132] Update the policy network parameters according to the display parameter update amount within the constraint space and the waterproof level update amount within the constraint space to obtain the updated display policy network and the updated waterproof policy network;

[0133] Input the updated display policy network and the updated waterproof policy network into the Actor network for forward calculation to obtain the display parameter action space and the waterproof level action space, and perform probability normalization on the display parameter action space and the waterproof level action space to obtain the display parameter adjustment probability distribution and the protection action probability distribution.

[0134] In this example, according to the display state value evaluation and the waterproof state value evaluation , calculate the ratio of the current policy and the old policy respectively, that is, the importance sampling ratio, which is expressed by the formula:

[0135] ;

[0136] Among them, is the probability distribution of the current policy taking action in state , represents the parameters of the policy network; and are the actions of display parameter adjustment and waterproof level control respectively. The importance sampling ratio is used to calculate the display parameter adjustment weight and the waterproof level adjustment weight respectively, which characterize the deviation degree of the current policy compared with the old policy. In order to limit the amplitude of policy update and avoid instability caused by too large or too small ratio, compare the display parameter adjustment weight and the waterproof level adjustment weight with the preset cropping range , where is a small positive number used to control the range of policy changes. The adjusted weights are truncated by the minimum value function, and the formula is:

[0137] ;

[0138] ;

[0139] where, and are the clipped display adjustment weight and the waterproof adjustment weight. The clipping operation ensures that the policy update remains within a safe range and prevents instability caused by large-scale updates. Based on the clipped weights, the policy objective function is calculated. The form of the policy objective function is to maximize the reward plus a regularization term, which is used to balance exploration and exploitation. The formula is expressed as:

[0140] ;

[0141] ;

[0142] where, and are the advantage functions of display and waterproofing, indicating the superiority of the current action compared to the average policy. The objective function is solved by a gradient optimization algorithm (such as the Adam optimizer) to obtain the display parameter policy gradient and the waterproof level policy gradient . The policy gradient is constrained to ensure that the update direction of the gradient does not exceed the preset range. The constraint mechanism is implemented by introducing a proximal mapping function, and the formula is expressed as:

[0143] ;

[0144] ;

[0145] where, is the maximum allowable value of the gradient, and are the gradient update directions of display and waterproofing. The update direction is projected into the constraint space through the proximal mapping function, and the display parameter update amount and the waterproof level update amount in the constraint space are calculated. The formula is:

[0146] ;

[0147] ;

[0148] where, is the learning rate, which is used to control the update step size. The policy network parameters are updated using the update amount to obtain the updated display policy network and waterproof policy network. The updated policy network is input into the Actor network for forward calculation to generate the display parameter action space and the waterproof level action space . To ensure that the sum of the action probabilities is 1, the action space is normalized probabilistically, and the formula is:

[0149] ;

[0150] The normalized action spaces respectively generate the display parameter adjustment probability distribution and the protection action probability distribution, providing a basis for the actual control of the device.

[0151] Referring to Figure 4 , this embodiment provides an intelligent environment sensing system based on an outdoor waterproof TV, including:

[0152] An acquisition module 401, configured to acquire an environmental monitoring data set of the outdoor waterproof TV through an environmental sensor array;

[0153] A feature extraction module 402, configured to input the environmental monitoring data set into an environmental sensing network for non-linear transformation and multi-dimensional feature extraction, and obtain a fused feature matrix through feature fusion;

[0154] A weighted calculation module 403, configured to calculate the temperature parameter weight coefficient, humidity parameter weight coefficient, light intensity parameter weight coefficient, and rainfall parameter weight coefficient for the fused feature matrix through a Bayesian optimization algorithm, and perform weighted calculation with the fused feature matrix to obtain an environmental risk factor;

[0155] A generation module 404, configured to input the environmental risk factor and the fused feature matrix into a deep neural network based on Proximal Policy Optimization (PPO) for policy iteration optimization calculation, and generate a protection control instruction set, where the protection control instruction set includes a display parameter adjustment instruction and a waterproof level control instruction;

[0156] An execution module 405, configured to respectively generate a driving signal for the display module and a control signal for the waterproof actuator according to the protection control instruction set, and execute the driving signal and the control signal through an execution unit.

[0157] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to that described in the above method embodiment, and details are not described herein again.

[0158] Referring to Figure 5 , this invention embodiment also provides a computer device, which can be a server, and its internal structure can be as Figure 5As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected via a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. The computer program, when executed by the processor, implements the above method.

[0159] Those skilled in the art can understand that Figure 5 the structure shown in is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0160] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0161] Those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. Among them, any reference to a memory, storage, database, or other medium provided by the present invention and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or an external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0162] It should be noted that, in this document, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article or method comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, apparatus, article or method. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of additional identical elements in the process, apparatus, article or method comprising such element.

[0163] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present invention.

Claims

1. An intelligent environment sensing method based on outdoor waterproof television, characterized in that: The following steps are involved: Collect environmental monitoring data sets of outdoor waterproof TVs through environmental sensor arrays; Inputting the environmental monitoring data set into the environmental sensor network for nonlinear transformation and multidimensional feature extraction, and obtaining a fused feature matrix through feature fusion; The temperature parameter weight coefficient, humidity parameter weight coefficient, light intensity parameter weight coefficient and rainfall parameter weight coefficient are calculated for the fusion feature matrix by using the Bayesian optimization algorithm, and weighted calculation is performed with the fusion feature matrix to obtain the environmental risk factor; Input the environmental risk factor and the fused feature matrix into the input layer of the deep neural network based on the proximal strategy optimization PPO, wherein the input layer includes 256 neurons, and performs nonlinear transformation through the ReLU activation function to obtain an initial feature vector; Input the initial feature vector into a shared layer in a deep neural network based on proximal strategy optimization PPO, wherein the shared layer comprises three fully connected layers, each layer comprises 128, 64, and 32 neurons respectively, and each neuron adopts a ReLU activation function to obtain a shared feature representation; The shared feature representation is input into the display parameter optimization branch and the waterproof level optimization branch respectively, wherein the display parameter optimization branch and the waterproof level optimization branch include 2 fully connected layers, each fully connected layer includes 16 neurons, and a ReLU activation function is used to obtain a display branch feature vector and a waterproof branch feature vector; Inputting the display branch feature vector and the waterproof branch feature vector into a value function network to perform state value calculation respectively, to obtain a display state value evaluation and a waterproof state value evaluation; The display state value evaluation and the waterproof state value evaluation are respectively input into a policy gradient calculation unit, and the policy gradient is calculated by an importance sampling ratio clipping method to obtain a display parameter adjustment probability distribution and a protection action probability distribution; Generate a display parameter adjustment strategy and a waterproof level control strategy based on the display parameter adjustment probability distribution and the protection action probability distribution, input the display parameter adjustment strategy and the waterproof level control strategy into an action mapping module, map discrete actions into continuous control quantities through a lookup table, and obtain display parameter adjustment instructions and waterproof level control instructions; Encapsulating the display parameter adjustment instruction and the waterproof level control instruction to generate a protection control instruction set; A driving signal of the display module and a control signal of the waterproof actuator are respectively generated according to the protection control instruction set, and the driving signal and the control signal are executed by the execution unit.

2. The intelligent environment sensing method based on outdoor waterproof television according to claim 1 is characterized in that: The environmental monitoring data set of the outdoor waterproof TV is collected by the environmental sensor array, including: Apply a preset sampling voltage to the input end of the temperature sensor in the environmental sensor array, perform analog signal amplification and filtering through the signal conditioning circuit, and obtain a temperature digital signal; An excitation signal is input to the detection end of the humidity sensor in the environmental sensor array, and impedance matching and signal stabilization are performed through a signal conversion circuit to obtain a humidity digital signal; Apply a reverse bias voltage to the photosensitive element of the light intensity sensor in the environment sensor array, perform current amplification and linearization processing through the photoelectric conversion circuit, and obtain a light intensity digital signal; Applying an AC excitation signal to the detection electrode of the rain sensor in the environmental sensor array, performing waveform shaping and digital processing through a signal processing circuit to obtain a rain digital signal; Input the temperature digital signal, the humidity digital signal, the light intensity digital signal and the rainfall digital signal into a data acquisition module, perform signal switching and sample-hold processing through a multiplexer, and obtain multi-channel sampling data; The multi-channel sampling data is subjected to digital quantity conversion and data compensation to obtain an environmental monitoring data set of an outdoor waterproof television.

3. The intelligent environment sensing method based on outdoor waterproof television according to claim 2 is characterized in that: The step of inputting the environmental monitoring data set into the environmental sensor network for nonlinear transformation and multidimensional feature extraction, and obtaining a fused feature matrix by feature fusion, includes: Inputting the environmental monitoring data set into the input layer of the environmental sensor network for data standardization and time series alignment to obtain a standardized data sequence; Input the standardized data sequence into the reservoir layer of the environmental sensor network for dynamic feature mapping, wherein the reservoir layer includes 128 neuron nodes, each neuron node uses a hyperbolic tangent function as an activation function, and the input weight matrix uses a sparse connection structure to obtain a reservoir state matrix; The state matrix of the storage tank is input into the time feature extraction unit, and the temperature change trend, humidity fluctuation characteristics, light intensity change pattern and rainfall accumulation characteristics are extracted through a recursive neural network structure to obtain a time series feature vector; Input the time series feature vector into the output layer of the environment sensor network for feature mapping to obtain an environment feature vector, and input the environment feature vector into a feature fusion network for feature enhancement to obtain an enhanced feature matrix; The enhanced feature matrix is ​​self-attention weighted combined to obtain an attention feature matrix, and the attention feature matrix is ​​dimensionally transformed to generate a fusion feature matrix.

4. The intelligent environment sensing method based on outdoor waterproof television according to claim 3 is characterized in that: The Bayesian optimization algorithm is used to calculate the temperature parameter weight coefficient, humidity parameter weight coefficient, light intensity parameter weight coefficient and rainfall parameter weight coefficient of the fused feature matrix, and weighted calculation is performed with the fused feature matrix to obtain the environmental risk factor, including: The objective function of the Bayesian optimization algorithm is constructed according to the fusion feature matrix, and the temperature parameter, humidity parameter, light intensity parameter and rainfall parameter are weighted spatially modeled by Gaussian process regression to obtain a joint distribution model of the four-dimensional parameters; The joint distribution model of the four-dimensional parameters is input into the Bayesian optimization algorithm, sampling points are selected based on the principle of maximizing expected lift, and the weight parameter space is iteratively sampled to obtain a weight parameter sample set; Calculating the sample mean and confidence interval according to the weight parameter sample set, and updating the weight coefficients by maximizing the posterior probability criterion to obtain the temperature parameter weight coefficient, the humidity parameter weight coefficient, the light intensity parameter weight coefficient and the rainfall parameter weight coefficient; Based on the temperature parameter weight coefficient, the humidity parameter weight coefficient, the light intensity parameter weight coefficient and the rainfall parameter weight coefficient, a multi-parameter weighted combination is performed on the fusion feature matrix to obtain a comprehensive scoring vector; The comprehensive score vector is hierarchically mapped through a piecewise linear mapping function to obtain an environmental risk factor.

5. The intelligent environment sensing method based on outdoor waterproof television according to claim 4 is characterized in that: The joint distribution model of the four-dimensional parameters is input into the Bayesian optimization algorithm, sampling points are selected based on the principle of maximizing expected lift, and the weight parameter space is iteratively sampled to obtain a weight parameter sample set, including: The joint distribution model of the four-dimensional parameters is input into a Gaussian kernel function, and the probability density value of each point in the parameter space is calculated by kernel function transformation to obtain a probability density distribution matrix; Performing parameter space grid division on the probability density distribution matrix to generate an initial sampling point set of four-dimensional parameters, and inputting the initial sampling point set into an expected lift calculation module to calculate the expected lift value of each sampling point to obtain an expected lift value sequence; Arrange the expected lift value sequence in descending order, select the first E points with the highest expected lift value as candidate sampling points, obtain a set of candidate sampling points, input the set of candidate sampling points into a local search module, perform gradient calculation around each candidate point with a preset search radius, and obtain a local gradient vector; The parameters of the candidate sampling points are updated according to the local gradient vector to obtain optimized sampling points, and the optimized sampling points are input into the Gaussian process regression model, the model parameters are updated and the posterior distribution is recalculated to obtain a weight parameter sample set.

6. The intelligent environment sensing method based on outdoor waterproof television according to claim 1, characterized in that: The display state value evaluation and the waterproof state value evaluation are respectively input into the policy gradient calculation unit, and the policy gradient is calculated by the importance sampling ratio clipping method to obtain the display parameter adjustment probability distribution and the protection action probability distribution, including: Based on the display status value evaluation and the waterproof status value evaluation, respectively calculating the new and old strategy ratios to obtain the display parameter adjustment weight and the waterproof level adjustment weight; Compare the display parameter adjustment weight and the waterproof level adjustment weight with a preset clipping range, perform truncation processing through a minimum value function, and obtain a clipped display adjustment weight and a clipped waterproof adjustment weight; Calculating a policy objective function based on the cropped display adjustment weight and the cropped waterproof adjustment weight to obtain a display parameter policy gradient and a waterproof level policy gradient, and constraining the display parameter policy gradient and the waterproof level policy gradient to obtain a display parameter update direction and a waterproof level update direction; Projecting the display parameter update direction and the waterproof level update direction through a proximal mapping function to obtain a display parameter update amount within the constraint space and a waterproof level update amount within the constraint space; The policy network parameters are updated according to the display parameter update amount in the constraint space and the waterproof level update amount in the constraint space to obtain an updated display policy network and an updated waterproof policy network; The updated display strategy network and the updated waterproof strategy network are input into the Actor network for forward calculation to obtain a display parameter action space and a waterproof level action space, and the display parameter action space and the waterproof level action space are probability normalized to obtain a display parameter adjustment probability distribution and a protection action probability distribution.

7. An intelligent environment sensing system based on outdoor waterproof television, characterized in that: The steps for implementing the intelligent environment sensing method based on an outdoor waterproof TV according to any one of claims 1 to 6, the intelligent environment sensing system based on an outdoor waterproof TV comprises: A collection module, used to collect environmental monitoring data sets of outdoor waterproof TVs through an environmental sensor array; A feature extraction module is used to input the environmental monitoring data set into the environmental sensor network for nonlinear transformation and multi-dimensional feature extraction, and obtain a fused feature matrix through feature fusion; A weighted calculation module, used to calculate the temperature parameter weight coefficient, humidity parameter weight coefficient, light intensity parameter weight coefficient and rainfall parameter weight coefficient for the fused feature matrix through a Bayesian optimization algorithm, and perform weighted calculation with the fused feature matrix to obtain an environmental risk factor; A generation module, used for inputting the environmental risk factor and the fusion feature matrix into a deep neural network based on proximal policy optimization (PPO) to perform policy iteration optimization calculation, and generating a protection control instruction set, wherein the protection control instruction set includes a display parameter adjustment instruction and a waterproof level control instruction; An execution module is used to generate a drive signal of the display module and a control signal of the waterproof actuator according to the protection control instruction set, and execute the drive signal and the control signal through an execution unit.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the intelligent environment sensing method based on an outdoor waterproof television as described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent environment sensing method based on an outdoor waterproof television according to any one of claims 1 to 6 are implemented.

Images