Intelligent control method for water conservancy facilities

Abstract

The present application relates to the technical field of water conservancy, in particular to a water conservancy facility sluice intelligent control method, the method comprises the following steps: step 1: the water body sparse spectrum data of target water area is collected; the water body sparse spectrum data is pretreated, and noise and spectrum distortion are eliminated; step 2: the spatial distribution image of target water area is obtained by multispectral imaging technology, and image decomposition is carried out, and the image gradient is obtained; the Fourier transform result is obtained by Fourier transform to key features; according to the image gradient and the Fourier transform result, the water level feature of target water area is extracted; step 3: a water level conversion model is established according to the water level feature of target water area extracted; water level prediction is carried out using the water level conversion model, and the water level prediction result is obtained; step 4: according to the water level prediction result, the operation of the water conservancy facility sluice is controlled. The water level measurement with wider range and more accurate precision is provided, the sluice control is carried out based on this, and the control accuracy is improved.

Description

Technical Field

[0001] The present disclosure relates to, but is not limited to, the field of water conservancy technology, and specifically to an intelligent control method for sluice gates of water conservancy facilities. Background Art

[0002] The effective management and utilization of water resources has always been a major challenge facing human society. Especially in regions with abundant water resources, such as rivers, lakes, and reservoirs, effective water level monitoring and flow control are required to ensure the smooth operation of various water resource utilization activities, including water supply, irrigation, flood control, and environmental protection. To achieve these goals, inventors have been working to improve the efficiency, accuracy, and intelligence of sluice control systems in water conservancy facilities.

[0003] Traditional sluice control systems typically rely on sensors, telemetry equipment, and manual operations to monitor and adjust water levels. While these systems can achieve water level control to a certain extent, they have several problems and limitations. First, traditional water level monitoring methods typically rely on a limited number of sensors, which can result in insufficient monitoring points to fully understand the conditions of the water area. Second, the collection and transmission of sensor data may be interfered with by environmental factors such as weather conditions and equipment failures, which may lead to inaccurate data. Furthermore, traditional water level control methods are typically based on empirical rules and manual intervention, lacking intelligence and automation, and are unable to cope with complex hydrological conditions and changes.

[0004] With the continuous advancement of science and technology, some advanced technologies have emerged in the field of water resources management, such as remote sensing, mathematical modeling, and multispectral imaging. However, the application of these technologies in water level monitoring and control still faces some challenges. For example, although traditional multispectral imaging technology can provide spatially distributed images, it usually requires a lot of computing resources and expertise to process and analyze the image data. In addition, existing multispectral imaging technology fails to fully utilize the correlation between spectral information and spatial information, resulting in the accuracy and stability of water level feature extraction to be improved. In addition, the prediction model of water level changes also requires more precise and intelligent methods to process time series data and error correlation to improve the accuracy of prediction. Summary of the Invention

[0005] The present disclosure provides an intelligent control method for sluice gates of water conservancy facilities, provides a water level measurement with a wider range and higher precision, and performs sluice gate control based on the water level measurement, thereby improving the accuracy of the control.

[0006] In order to solve the above problems, the technical solution of the present invention is achieved as follows:

[0007] A method for intelligently controlling a sluice gate of a water conservancy facility, the method comprising:

[0008] Step 1: Collect sparse spectral data of the target water area; preprocess the sparse spectral data of the water body to eliminate noise and spectral distortion; use the sparse coding method to decompose the preprocessed sparse spectral data of the water body and extract key features;

[0009] Step 2: Obtain the spatial distribution image of the target water area by multi-spectral imaging technology, and perform image decomposition to obtain the image gradient; perform Fourier transform on the key features to obtain the Fourier transform result; extract the water level features of the target water area according to the image gradient and the Fourier transform result;

[0010] Step 3: Establish a water level transformation model according to the extracted water level features of the target water area; use the water level transformation model to predict the water level and obtain the water level prediction result;

[0011] Step 4: Control the operation of the water conservancy facility sluice according to the water level prediction result.

[0012] Further, in step 1, the collected sparse spectral data of the target water area is represented by the following formula:

[0013] D(λ,t)=I0(λ,t)*e ―σ(λ)N(t) *R(λ,t);

[0014] Where D(λ,t) is the time-varying detected sparse spectral data of the water body, I0(λ,t) is the time-varying incident light spectrum, σ(λ) is the wavelength-dependent absorption cross-section, N(t) is the molecular number density of the atmospheric column, and R(λ,t) is the reflectivity of the water body of the target water area.

[0015] Further, in step 1, the sparse spectral data of the water body is preprocessed using the following formula:

[0016] Where D'(λ,t) is the preprocessed sparse spectral data of the water body, B(λ,t) is the time and wavelength dependent baseline spectrum, S(λ,t) is the system response function, and L(λ,t) is the spectral smoothing function.

[0017] Further, in step 1, the sparse coding method is used to decompose the preprocessed sparse spectral data of the water body and extract key features, which is to find a coefficient matrix X(t) that minimizes the reconstruction error, meets the sparsity requirement, and has smooth time variation, which is represented by the following formula:

[0018]

[0019] Where A(λ, t) is a known basis matrix whose columns represent different spectral bases; X(t) is a coefficient matrix representing key features, and each column corresponds to a time point t, describing how to linearly combine the bases in the basis matrix A at that time point to reconstruct D′(λ, t); ||·||2 represents the L2 norm, which is used to measure the smoothness of the reconstruction error and coefficient changes; ||·||1 represents the L1 norm, which is used to increase the sparsity of the coefficient matrix X(t); λ and γ are both regularization parameters, which are set values ​​used to control the influence of the L1 regularization term and the smoothing term on the entire formula. A higher λ value will result in a sparser X(t), while a higher γ value will result in an X(t) that changes more smoothly over time. It is used to measure the degree of change of the coefficient matrix X(t) over time in order to find an X(t) that changes smoothly over time.

[0020] Furthermore, the spatial distribution image of the target water area obtained by multispectral imaging technology in step 2 is expressed using the following formula:

[0021]

[0022] Among them, G(x, y, t) is the spatial distribution image of the target water area acquired at time t, which is a two-dimensional image; F(x―x′, y―y′, t) represents the radiance distribution function of the target water area; H(x′, y′, t) is the point spread function of multispectral imaging, which describes the response of multispectral imaging to a point light source; η(x, y, t) is the noise term, which represents the uncertainty and error introduced in the multispectral imaging process; x is the position coordinate of the x-axis in the target water area, and y is the position coordinate of the y-axis in the target water area; x′ is the change value of the x-axis; y′ is the change value of the y-axis; the integral in ∫∫F(x―x′, y―y′, t)H(x′, y′, t)dx′dy′ is performed with respect to x′ and y′, indicating that the convolution of F and H is performed on the plane of the entire spatial distribution image; is the convolution operator symbol.

[0023] Furthermore, the method for performing image decomposition on the spatial distribution image in step 2 includes: using the conjugate gradient method to solve the following optimization problem:

[0024]

[0025] V is the matrix corresponding to the spatial distribution image, which contains all the pixel values ​​of the entire spatial distribution image; the goal of this optimization problem is to find the best W and H to minimize the reconstruction error while maintaining the sparsity of W and preserving the intrinsic structure of the spatial distribution image; W is the basis image matrix; H is the coefficient matrix; ||.|| F is the Frobenius norm; ||.|| 2，1is a group sparsity regularization term used to promote the sparsity of the basis matrix W; tr(.) represents the trace of the matrix, that is, the sum of the diagonal elements of the matrix; L is the Laplace matrix; α and β are regularization parameters and are set values.

[0026] Furthermore, the following formula is used in step 2 to extract the water level characteristics of the target water area based on the image gradient and Fourier transform results:

[0027]

[0028] in, is the water level characteristic of the target water area, which is a time-dependent characteristic vector; is the gradient of the decomposed spatial distribution image, is the Fourier transform result after processing, and ψ is the feature extraction function.

[0029] Furthermore, the feature extraction function ψ is a composite function, which is derived from and Extract features from Extract features and obtain the gradient features of the spatial distribution image, which reflects the average rate of change of the gradient of the spatial distribution image; Find the frequency domain extraction features in and obtain the frequency domain features of the sparse spectral data of the water body;

[0030] Furthermore, in step 3, a water level transformation model is established based on the extracted water level characteristics of the target water area, and is expressed using the following formula:

[0031]

[0032] Among them, y(t) is the water level prediction result, β0 and β t is the regression coefficient, κ(t―t′) is the autocorrelation function, ∈(t′) is the error term; t′ is the time variation.

[0033] The intelligent control method for sluice gates of water conservancy facilities of the present invention has the following beneficial effects: Traditional water level monitoring usually relies on a limited number of water level sensors, which are distributed at different monitoring points, resulting in limited monitoring accuracy. In contrast, the present invention uses multispectral imaging technology to obtain spatial distribution images of the target water area in real time, providing more comprehensive and detailed water level monitoring data. This method can capture small fluctuations in water level changes throughout the water area, thereby improving the accuracy of monitoring. Traditional water level control methods are usually based on empirical rules and manual intervention and lack intelligence. In contrast, the present invention can achieve automated water level control through data processing and water level prediction models. The system can intelligently adjust the operation of the sluice gate based on real-time monitoring data and prediction results to meet the needs of different water resource utilization activities. This helps to improve the intelligence level of water conservancy facilities, reduce manual intervention, and lower operational risks. Since traditional water level sensors are placed in water for a long time, temperature changes can affect the density and viscosity of the liquid, thereby affecting the performance of the water level sensor. The liquid may contain corrosive substances or pollutants, which may damage the external or internal components of the sensor and reduce its performance or life. The present invention can avoid this situation. At the same time, the present invention can also perform water level measurement in some places where it is inconvenient to install water level sensors. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of a method flow chart for an intelligent control method of a water gate in a water conservancy facility provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present disclosure more clear and understandable, the present disclosure is 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 disclosure and are not intended to limit the present disclosure.

[0036] refer to Figure 1 , a water conservancy facility sluice intelligent control method, the method comprising:

[0037] Step 1: Collect sparse spectral data of the target water area; preprocess the sparse spectral data of the water body to eliminate noise and spectral distortion; use sparse coding method to decompose the preprocessed sparse spectral data of the water body and extract key features;

[0038] The first task in this step is to obtain sparse spectral data of the target water body. This data can be obtained through various means, such as remote sensing satellites, drones, sensor networks, or ground-based measurement instruments. This data typically includes spectral reflectance or absorption information measured across different wavelength ranges. The obtained sparse spectral data of water bodies may be affected by various factors, including atmospheric disturbances, cloud cover, and impurities in the water. Therefore, preprocessing is required to ensure data accuracy and reliability.

[0039] The preprocessing process generally includes the following steps: Noise removal: Filtering techniques or signal processing methods are used to eliminate random noise in the data to improve data quality. Spectral correction: Corrections are performed to correct distortions in the spectral data, such as atmospheric correction and inter-band correction, to ensure consistency of the data across time and location. Brightness and temperature correction: Based on the characteristics of the sensor, the data is corrected for brightness and temperature to eliminate the effects of lighting conditions and temperature changes on the data. Background subtraction: Based on the ambient spectrum, the background signal is subtracted from the measured spectral data to highlight the spectral characteristics of the water body.

[0040] The sparse coding method decomposes the preprocessed data:

[0041] Sparse coding is a signal processing technique that aims to represent high-dimensional data in a lower-dimensional form while retaining key information. In this step, the preprocessed sparse spectral data of the water body is decomposed into a series of basic spectral features. The sparse coding method usually involves representing the original data as a sparse vector in which only a few elements are non-zero. These non-zero elements correspond to contributions to specific features. Through sparse coding, key features in the target water area can be extracted. These features may include information such as water quality, water depth, and dissolved substance concentration, which are of great significance in the intelligent control of sluices.

[0042] Step 2: Obtain the spatial distribution image of the target water area through multispectral imaging technology, and perform image decomposition on it to obtain image gradients; perform Fourier transform on key features to obtain Fourier transform results; and extract the water level characteristics of the target water area based on the image gradients and Fourier transform results;

[0043] Multispectral imaging is a remote sensing technology that uses spectral sensors with multiple wavelengths to capture images of the surface or water. These wavelengths typically fall within the visible and infrared spectrums. In the context of a water conservancy facility's sluice gates, multispectral imaging might use satellites, drones, or other image acquisition devices to acquire image data of the target water area. This image data contains reflectance information of the water area at different wavelengths.

[0044] After acquiring multispectral images of a body of water, these images can be decomposed to identify and extract information about the water's characteristics. Image decomposition may include the use of image processing techniques such as filtering and edge detection. Image gradient refers to the rate of change between different regions in an image. In the context of water conservancy facilities, image gradients can help identify areas of varying water levels and flow rates within a body of water. By calculating image gradients, areas of varying water levels can be identified, thereby extracting water level characteristics.

[0045] The Fourier transform is a mathematical tool used to convert signals or data from the time domain to the frequency domain. In this step, the extracted water level feature data is usually time domain data, and the Fourier transform can convert it into a frequency domain representation. The result of the Fourier transform is a spectrum diagram that shows the contribution of different frequency components. In the context of water bodies, this can be used to analyze periodic changes in water levels, such as tidal cycles or other hydrological periodicities. By combining image gradient information and Fourier transform results, water level characteristics of the target water body can be extracted. These characteristics include water level changes, fluctuations, and possible tidal or flow information. The extracted water level features can be used for further analysis and water level prediction to assist in the intelligent control of sluice gate operations.

[0046] Step 3: Establish a water level transformation model based on the extracted water level characteristics of the target water area; use the water level transformation model to predict the water level and obtain the water level prediction result;

[0047] Step 4: Control the operation of the sluice gates of water conservancy facilities based on the water level prediction results.

[0048] Preferably, in step 1, the collected sparse spectral data of the target water area is expressed using the following formula:

[0049] D(λ,t)=I0(λ,t)*e ―σ(λ)N(t) *R(λ,t);

[0050] Where D(λ,t) is the time-varying detected sparse spectral data of water bodies, I0(λ,t) is the time-varying incident light spectrum, σ(λ) is the wavelength-dependent absorption cross section, N(t) is the molecular number density of the atmospheric column, and R(λ,t) is the reflectance of the water body in the target water area.

[0051] Specifically, D(λ, t) is the time-varying sparse spectral data detected for water. It represents the light intensity or radiation intensity detected after light passes through the water at a specific wavelength (λ) and time (t). This data is obtained from the water and contains information about the water's absorption and reflection of light. I0(λ, t) is the time-varying incident light spectrum. It represents the light intensity or radiation intensity of light incident on the water at the same wavelength (λ) and time (t). The properties of the incident light may vary over time; for example, the spectral characteristics of sunlight may differ at different times and under different weather conditions. σ(λ) is the wavelength-dependent absorption cross section. It describes the absorption rate of water at different wavelengths. σ(λ) is a function whose value varies with wavelength λ. Different wavelengths of light are absorbed differently in water, which is related to the optical properties of the substances present in the water. Longer wavelengths of light may be more easily absorbed by water, while shorter wavelengths may pass through the water. N(t): This is the molecular number density of the atmospheric column, representing the number of molecules in the atmosphere at a specific time t. The number of molecules in the atmosphere can affect the propagation and scattering of light in the atmosphere. This term represents the atmospheric effect on light propagation. R(λ,t): This is the reflectivity of the water body in the target area. It indicates the degree to which the water body reflects incident light at a specific wavelength λ and time t. Reflectivity is typically a value between 0 and 1, indicating the proportion of light that is reflected. The reflectivity of water bodies is affected by many factors, including the color, turbidity, and surface shape of the water. The formula describes how the detected spectral data (D(λ,t)) is determined by the incident light spectrum (I0(λ,t)), the absorption cross section (σ(λ)), the molecular number density of the atmospheric column (N(t)), and the reflectivity of the water body (R(λ,t)).

[0052] Preferably, in step 1, the following formula is used to preprocess the sparse spectral data of water:

[0053]

[0054] Where D′(λ,t) is the preprocessed sparse spectral data of water, B(λ,t) is the time- and wavelength-dependent baseline spectrum, S(λ,t) is the system response function, and L(λ,t) is the spectral smoothing function.

[0055] Specifically, D′(λ, t) is preprocessed water sparse spectral data, representing spectral data at a specific wavelength (λ) and time (t), obtained after a series of processing steps. This preprocessed data may be used for subsequent analysis and modeling. D(λ, t) is the raw, detected water sparse spectral data, which contains information about the water's absorption and reflectance. B(λ, t) is the time- and wavelength-dependent baseline spectrum. The baseline spectrum typically represents the background or noise component in the spectral data. Subtracting the baseline spectrum from the raw data can remove non-target signals and improve the signal-to-noise ratio. S(λ, t) is the system response function, which describes the response characteristics of the spectral instrument or sensor. Different instruments may have different response functions, so this term needs to be considered when correcting the data to obtain accurate water spectral information. L(λ, t) is the spectral smoothing function, used to smooth the data. Spectral smoothing can help remove high-frequency noise or fluctuations, making the data more stable and analyzable. Common smoothing methods include moving average and Gaussian smoothing.

[0056] The entire formula first subtracts the baseline spectrum from the raw spectral data to remove background and noise. Then, the system response function is considered to correct for instrument characteristics and ensure data accuracy. Finally, a spectral smoothing function is applied to make the data more readable and analyzable.

[0057] Preferably, in step 1, the sparse coding method is used to decompose the preprocessed water sparse spectral data. The method for extracting key features is to find a coefficient matrix X(t) that minimizes the reconstruction error, meets the sparsity requirement, and has a smooth time variation. The process is expressed by the following formula:

[0058]

[0059] Where A(λ, t) is a known basis matrix whose columns represent different spectral bases; X(t) is a coefficient matrix representing key features, and each column corresponds to a time point t, describing how to linearly combine the bases in the basis matrix A at that time point to reconstruct D′(λ, t); ||·||2 represents the L2 norm, which is used to measure the smoothness of the reconstruction error and coefficient changes; ||·||1 represents the L1 norm, which is used to increase the sparsity of the coefficient matrix X(t); λ and γ are both regularization parameters, which are set values ​​used to control the influence of the L1 regularization term and the smoothing term on the entire formula. A higher λ value will result in a sparser X(t), while a higher γ value will result in an X(t) that changes more smoothly over time. It is used to measure the degree of change of the coefficient matrix X(t) over time in order to find an X(t) that changes smoothly over time.

[0060] Specifically, the reconstruction error term The role of this term is to measure the error between the data reconstructed by linearly combining the basis matrix A(λ, t) with the coefficient matrix X(t) and the pre-processed original data D'(λ, t). By adjusting the coefficient matrix X(t), we try to minimize this error so that the result of linear combination is as close to the original data as possible. This helps to extract information about water body characteristics from the original data. The L1 regularization term λ|X(t)|1: The role of the L1 regularization term is to introduce sparsity, encouraging most elements in the coefficient matrix X(t) to be zero. By increasing the L1 regularization term, we force the coefficient matrix X(t) to become sparse, i.e., only a few important features will be retained in the coefficient matrix. This helps to reduce the dimensionality of the data, improve computational efficiency, and help identify the main water body characteristics. The smoothing term The role of the smoothing term is to ensure that the changes in the coefficient matrix X(t) over time are smooth, i.e., the coefficient matrix does not fluctuate dramatically in a short period of time. By introducing the smoothing term, we perform L2 regularization on the derivative of the coefficient matrix to prevent excessive fluctuations. This helps to capture gradual changes in water body characteristics rather than instability caused by noise or short-term fluctuations. λ and γ are regularization parameters, whose values can be adjusted according to specific application requirements. A higher λ value will result in stronger sparsity, i.e., more coefficients being zero, thus reducing the number of features. A higher γ value will result in a smoother coefficient matrix X(t), which helps to capture gradual changes in characteristics.

[0061] In summary, the principle of this formula is to find a coefficient matrix X(t), where each column corresponds to a time point t, describing how to linearly combine the bases in the basis matrix A to reconstruct the pre-processed water body sparse spectral data D'(λ, t). By this method, we can extract key water body characteristics from the original data for subsequent analysis, modeling or decision-making.

[0062] Preferably, the spatial distribution image of the target water area obtained by the multi-spectral imaging technology in step 2 is expressed by the following formula:

[0063]

[0064] Where G(x, y, t) is a spatially distributed image of the target water body acquired at time t, F(x-x', y-y', t) represents the radiance distribution function of the target water body, H(x', y', t) is the point spread function of the multispectral imaging, describing the response of the multispectral imaging to a point light source, η(x, y, t) is a noise term, representing the uncertainty and error introduced in the process of multispectral imaging, x is the position coordinate of the x-axis in the target water body, y is the position coordinate of the y-axis in the target water body, x' is the change value of the x-axis, y' is the change value of the y-axis, and the integral in ∫∫F(x-x', y-y', t)H(x', y', t)dx' dy' is performed with respect to x' and y', indicating that the convolution of F and H is performed on the plane of the entire spatially distributed image. is a convolution operator.

[0065] Specifically, the radiance of the target scene (in this case, the water body) is captured using an imaging sensor (such as a camera, spectrometer, etc.). This typically depends on the physical properties of the sensor (e.g., pixel size, spectral response, dynamic range, etc.) and environmental factors (such as lighting, weather conditions, etc.). The scene radiance function F(x, y, t) describes the scene radiance at different positions (x, y) and time t. The imaging system (including the sensor and possibly optical elements) has its own response characteristics to the captured radiance, which is usually described as a point spread function (PSF) H(x, y, t). The PSF defines how an ideal point source propagates through the imaging system and forms an image on the sensor. By integrating the scene radiance and the response of the imaging system, the resulting image can be simulated through a convolution process involving the scene radiance function F(x, y, t) and the point spread function H(x, y, t). Therefore, the acquired image G(x, y, t) is the convolution of F and H, plus a possible noise term η(x, y, t), which represents system noise and / or environmental noise.

[0066] G(x, y, t): This is the spatial distribution image of the target water area. It represents the distribution of radiant or reflective intensity at different locations (x, y) at time t. This image contains spatial information about the water body and can be used to analyze its properties and characteristics. F(x-x′, y-y′, t): This is the radiance distribution function of the target water area. It describes the radiant properties of the water body at different locations (x, y) and time t, that is, how the water reflects or emits light. This function plays a key role in multispectral imaging, determining the brightness at different locations in the final image. H(x′, y′, t): This is the point spread function of multispectral imaging, describing the response of the multispectral imaging system to a point light source. It includes the characteristics of the optical system, such as the lens and sensor, as well as the diffusion and propagation effects of light during propagation. The point spread function takes into account the engineering characteristics of the imaging system, which affects the image blur and spatial resolution. η(x, y, t): This is the noise term, representing the uncertainty and error introduced in the multispectral imaging process. In actual imaging, there may be various noise sources, including electronic noise and environmental noise. This term represents the noise component in the image and affects the image quality. ∫∫F(x―x′,y―y′,t)H(x′,y′,t)dx′dy′: This is a convolution operation, which represents the convolution operation of the radiance distribution function F and the point spread function H on the entire spatial distribution image plane. This convolution operation simulates the propagation process of light from the water body to the imaging system. The entire formula simulates the propagation and reflection of light in the water body, and then combines the radiance distribution F with the point spread function H through the convolution operation to obtain the spatial distribution image G(x,y,t) of the water body at time t. This image contains the reflective properties of the water body, spatial distribution information, and the response characteristics of the imaging system. Finally, the noise term η is used to account for the uncertainty and error in the imaging process.

[0067] Preferably, the method of performing image decomposition on the spatial distribution image in step 2 includes: using the conjugate gradient method to solve the following optimization problem:

[0068]

[0069] V is the matrix corresponding to the spatial distribution image, which contains all the pixel values ​​of the entire spatial distribution image; the goal of this optimization problem is to find the best W and H to minimize the reconstruction error while maintaining the sparsity of W and preserving the intrinsic structure of the spatial distribution image; W is the basis image matrix; H is the coefficient matrix; ||.|| F is the Frobenius norm; ||.|| 2,1 is a group sparsity regularization term used to promote the sparsity of the basis matrix W; tr(.) represents the trace of the matrix, that is, the sum of the diagonal elements of the matrix; L is the Laplace matrix; α and β are regularization parameters and are set values.

[0070] Basis matrix W: This matrix contains a series of basis images or patterns, which are the key features extracted from the original image V. Each basis (i.e., a column of W) can be regarded as a component of the original image. In different application scenarios, these bases may correspond to edges, textures, color distributions or other meaningful image features. In the context of water level monitoring, some bases may specifically correspond to visual properties of the water surface, such as reflection, color or ripples. Coefficient matrix H: Each column of this matrix provides the weight of the basis corresponding to an original image (if V is an image set or a series of images). These weights indicate the degree of contribution of each basis when reconstructing the original image. In other words, H describes how to combine the bases in the basis matrix W to best approximate the original image data.

[0071] Image decomposition can decompose a complex image or image sequence into a set of simpler elements. This not only reduces the complexity and storage requirements of the data, but also makes subsequent image analysis tasks (such as feature extraction, classification, anomaly detection, etc.) more efficient and accurate. In the application of water level monitoring, by analyzing the changes in the coefficient matrix H, it may be possible to detect changes in the water level. V: This is the spatial distribution image to be decomposed, which contains the numerical information of each pixel in the image. This image may contain information such as the reflection or emission intensity distribution of the water body. W and H: These are the two matrices to be solved. W is the basis image matrix, which contains a set of basis images and can be regarded as a set of basic image patterns. H is the coefficient matrix, which contains the weight information of each basis image in the image. By multiplying the basis image with the coefficient matrix, the original image can be reconstructed.

[0072] Reconstruction error term The goal of this term is to minimize the difference between the decomposed image and the original image. It uses the Frobenius norm to measure the distance between the two matrices. By adjusting W and H, it attempts to minimize this error to get as close to the original image as possible.

[0073] The group sparsity regularization term α|W| 2,1 : The goal of this term is to promote the sparsity of the basis image matrix W. |W| 2,1 Represents applying the L2 norm to each column of W and applying the L1 norm to the L2 norm of these columns. This regularization term encourages most basis images to be zero, thus achieving sparsity in the basis.

[0074] Smoothness regularization term βtr(HLH TThe goal of this term is to preserve the intrinsic structure of the image and promote the smoothness of the coefficient matrix H in the image space. Through the trace operation tr(.), it considers the smoothness of the coefficient matrix H in order to preserve the relationship between pixels in the image.

[0075] The goal of the entire optimization problem is to find the optimal basis image matrix W and coefficient matrix H to minimize the reconstruction error, promote the sparsity of the basis, and preserve the intrinsic structure of the image. This decomposition method is commonly used in image processing and machine learning for tasks such as feature extraction, denoising, image compression, etc., because it can represent the image as a linear combination of the basis, thus better understanding and processing image data. At the same time, the introduction of the regularization term helps to control the complexity of the model and avoid overfitting problems.

[0076] Preferably, the following formula is used in step 2 to extract the water level features of the target water area according to the image gradient and Fourier transform results:

[0077]

[0078] where, is the water level feature of the target water area, which is a time-dependent feature vector; is the gradient of the decomposed spatial distribution image, is the processed Fourier transform result, and ψ is the feature extraction function.

[0079] Gradient information Gradient refers to the rate of change of pixel values in an image. In this formula, represents the gradient of the decomposed spatial distribution image. In the water area, the change of water level usually leads to the change of pixel values in the image. By calculating the gradient of the image, the location and amplitude of the water level change in the image can be captured. This information is very important for the extraction of water level features. Fourier transform Fourier transform is used to convert a signal from time domain to frequency domain. In this formula, represents the processed Fourier transform result. Fourier transform can analyze the frequency components in the signal, which is also useful for water level feature extraction. The periodic changes in water level can exhibit specific frequency components in the frequency domain, and through Fourier transform, these frequency components can be extracted. Feature extraction function (ψ): The role of the feature extraction function ψ is to combine the gradient information and the Fourier transform result to extract the water level features of the target water area. This function can include various operations such as specific filtering, frequency domain analysis, feature selection, etc. Its design depends on the specific requirements of the task. For example, if the water level change is related to frequency, ψ may focus on certain frequency components in the frequency domain, and if the water level change is related to gradient, ψ may extract specific features based on gradient information. Water level feature vector Finally, by processing the gradient information and the Fourier transform result through the feature extraction function ψ, the water level feature vector This feature vector contains information about the trend of water level changes over time, frequency components, and possibly spatial distribution information. It is a time-dependent vector that can be used for monitoring water level changes, analyzing water body dynamics, and other applications.

[0080] Preferably, the feature extraction function ψ is a composite function that extracts features from and By extracting features from , the gradient feature of the spatial distribution image is obtained, which reflects the average rate of change of the gradient of the spatial distribution image; by finding the frequency domain extraction feature from , the frequency domain feature of the water body sparse spectral data is obtained.

[0081] Extracting features from : represents the gradient of the decomposed spatial distribution image. This gradient contains the rate of change information of each pixel position in the image. The feature extraction function extracts features from , which may involve the following operations: calculating the average rate of change of the gradient: this can be achieved by calculating the average value or other statistical quantities of the gradient values. This feature reflects the overall trend of the gradient of the spatial distribution image, which can help capture the average speed and direction of water level changes in the image. represents the frequency domain result of the water body sparse spectral data. This frequency domain result contains the representation of the spectral data in the frequency domain, reflecting the existence and intensity of different frequency components. The feature extraction function extracts features from Feature extraction, which can include the following operations: frequency domain analysis: this can include filtering the frequency spectrum, finding the main frequency components, calculating the energy distribution of the frequency domain features, etc. These features can help capture the frequency information in the sparse spectrum data of the water body, and help understand the periodic characteristics of water level changes.

[0082] Preferably, in step 3, the water level transformation model is established according to the extracted water level characteristics of the target water area, and the following formula is used to represent it:

[0083]

[0084] Where y(t) is the water level prediction result, β0 and β t are regression coefficients, κ(t―t') is the autocorrelation function, ∈(t') is the error term; t' is the time change.

[0085] Specifically, y(t): This is the result of water level prediction, representing the predicted water level value at time t. This is the target that we hope to predict through the model. β0 and β t : These are regression coefficients, used to linearly combine the components of water level features and the constant term β0, thus establishing the prediction model of water level. They represent the weight of each feature in water level prediction. This part represents the linear combination of water level features , where each feature is multiplied by the corresponding regression coefficient β t and then summed. This part simulates the contribution of water level features to water level prediction. ∫k(t―t')∈(t')dt': This is an integral term, representing the integration of error term ∈(t') through autocorrelation function κ(t―t'). This part considers the correlation of errors over time to capture the time correlation in water level prediction. κ(t―t'): This is the autocorrelation function, which is used to measure the correlation of errors over time. Autocorrelation function describes the correlation between errors at different time points, which can help consider the propagation and influence of errors over time. ∈(t'): This is the error term, representing the part of the model that cannot perfectly predict the water level. These errors usually include uncertainties due to model simplification, measurement errors or unconsidered factors.

[0086] The principle of the whole formula is to establish a water level transformation model to predict water level changes by linearly combining water level features and considering the error correlation over time. Regression coefficients β0 and β tThis can be obtained through model training. The autocorrelation function κ(t-t′) can be determined based on data analysis, while the error term ∈(t′) typically includes variations and uncertainties that the model cannot perfectly capture. The goal of this model is to predict water level trends based on water level characteristics and temporal error correlations, thereby enabling water level monitoring and forecasting. The model's performance typically requires verification and adjustment using real-world data to ensure it accurately captures dynamic water level changes.

[0087] Those skilled in the art will appreciate that all or some of the steps in the methods, systems, and functional modules / units in the apparatus disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0088] The preferred embodiments of the present disclosure are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present disclosure. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present disclosure shall fall within the scope of the present disclosure.

Claims

1. A method for intelligently controlling a sluice gate of a water conservancy facility, characterized in that: The method comprises: Step 1: Collect sparse spectral data of the target water area; preprocess the sparse spectral data of the water body to eliminate noise and spectral distortion; use sparse coding method to decompose the preprocessed sparse spectral data of the water body and extract key features; Step 2: Obtain the spatial distribution image of the target water area through multispectral imaging technology, and perform image decomposition on it to obtain image gradients; perform Fourier transform on key features to obtain Fourier transform results; and extract the water level characteristics of the target water area based on the image gradients and Fourier transform results; Step 3: Establish a water level transformation model based on the extracted water level characteristics of the target water area; use the water level transformation model to predict the water level and obtain the water level prediction result; Step 4: Control the operation of the sluice gates of the water conservancy facilities based on the water level prediction results; In step 1, the sparse spectral data of the target water area collected is expressed using the following formula: ； in, is the time-varying detected sparse spectral data of water, is the time-varying incident light spectrum, is the wavelength-dependent absorption cross section, is the molecular number density of the atmospheric column, is the reflectivity of the water body in the target water area; In step 1, the following formula is used to preprocess the sparse spectral data of water bodies: ； in, is the preprocessed sparse spectral data of water, is the time- and wavelength-dependent baseline spectrum, is the system response function, is the spectral smoothing function; In step 1, the sparse coding method is used to decompose the preprocessed water sparse spectral data. The method for extracting key features is to find a coefficient matrix , which can minimize the reconstruction error, meet the sparsity requirements, and its time variation is smooth. The process is expressed using the following formula: ； in, is a known basis matrix whose columns represent different spectral bases; Is a coefficient matrix, representing the key features, each column corresponds to a time point , which describes how to linearly combine the basis matrices at that point in time Reconstruction of the base ; Represents the L2 norm, which is used to measure the smoothness of reconstruction error and coefficient changes; Represents the L1 norm, used to increase the coefficient matrix Sparsity; and Both are regularization parameters, which are set values ​​to control the influence of L1 regularization term and smoothing term on the whole formula. Values ​​will result in a sparser , and higher Values ​​of ; Used to measure the coefficient matrix The degree of change over time, in order to find a smooth time-varying .

2. The intelligent control method for water conservancy facility sluice gates according to claim 1, characterized in that: In step 2, the spatial distribution image of the target water area obtained by multispectral imaging technology is expressed using the following formula: ； in, It's in time The acquired spatial distribution image of the target water area is a two-dimensional image; Represents the radiometric distribution function of the target water area; is the point spread function of multispectral imaging, which describes the response of multispectral imaging to a point light source; is the noise term, representing the uncertainty and error introduced in the multispectral imaging process; For the target waters The position coordinates of the axis, For the target waters The position coordinates of the axis; for The change value of the axis; for The change value of the axis; The integral in is about and carried out, indicating and The convolution is performed on the plane of the entire spatially distributed image; is the convolution operator symbol.

3. The intelligent control method for water conservancy facility sluice gate according to claim 2, characterized in that: The method for performing image decomposition on the spatial distribution image in step 2 includes using the conjugate gradient method to solve the following optimization problem: ； is the matrix corresponding to the spatial distribution image, which contains all the pixel values ​​of the entire spatial distribution image; the goal of this optimization problem is to find the best and , minimizing the reconstruction error while maintaining The sparsity of ,and preserve the intrinsic structure of spatially distributed images; is the basis image matrix; is the coefficient matrix; is the Frobenius norm; is a regularization term for group sparsity, which is used to promote the basis matrix Sparsity; represents the trace of the matrix, that is, the sum of the diagonal elements of the matrix; is the Laplace matrix; and are regularization parameters and are set values.

4. The intelligent control method for water conservancy facility sluice gates according to claim 3, characterized in that: In step 2, the following formula is used to extract the water level characteristics of the target water area based on the image gradient and Fourier transform results: ； in, is the water level characteristic of the target water area, which is a time-dependent characteristic vector; is the gradient of the decomposed spatial distribution image, is the Fourier transform result after processing, is the feature extraction function.

5. The intelligent control method for water gates of water conservancy facilities according to claim 4, characterized in that: The feature extraction function is a composite function, which is derived from and Extract features from By Extract features and obtain the gradient features of the spatial distribution image, which reflects the average rate of change of the gradient of the spatial distribution image; The frequency domain extraction features are found in the ,frequency domain features of the sparse spectral data of water bodies are obtained.

6. The intelligent control method for water gates of water conservancy facilities according to claim 5, characterized in that: In step 3, a water level transformation model is established based on the extracted water level characteristics of the target water area, and is expressed using the following formula: ； in, is the water level prediction result, and is the regression coefficient, is the autocorrelation function, is the error term; is the time variation.

Images