A Machine Learning-Based Method for Dynamic Monitoring of Reservoir Water Levels

Abstract

This invention relates to the field of data processing technology, and in particular to a method for dynamic monitoring of reservoir water levels based on machine learning. This method acquires multidimensional monitoring data of the reservoir at each monitoring moment within a preset historical period prior to the current monitoring moment, constructs a three-dimensional spatial coordinate system, fits the data points in the three-dimensional spatial coordinate system to obtain a fitted straight line, and filters at least one suspicious environmental time interval within the preset historical period based on the distance difference between each data point in the three-dimensional spatial coordinate system and the fitted straight line. Based on the fluctuation characteristics of the multidimensional monitoring data within each suspicious environmental time interval, the multidimensional monitoring data within each suspicious environmental time interval is classified to obtain the data classification result corresponding to each suspicious environmental time interval. This result is used to adjust the preset baseline learning rate in the GBDT model to obtain an adaptive learning rate. Based on the adaptive learning rate, an optimized model is obtained, improving the accuracy of dynamic monitoring of reservoir water levels.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for dynamic monitoring of reservoir water levels based on machine learning. Background Technology

[0002] As a core component of water conservancy infrastructure, the safe operation of reservoirs is directly related to flood control, water supply security, and ecological balance within a watershed. Reservoir water level, a key indicator of reservoir operation, is influenced by a combination of factors including precipitation, runoff, dispatching operations, and climate fluctuations. Therefore, dynamic monitoring of reservoir water levels is of great significance for ensuring the safe and stable operation of reservoirs and improving the intelligence level of water conservancy dispatching.

[0003] Existing methods typically utilize Gradient Boosting Decision Tree (GBDT) models for dynamic monitoring of reservoir water levels. First, initial predictions are obtained based on weak learners, and residuals are calculated. Then, environmental factors such as temperature and humidity are incorporated, and different weights are assigned along the gradient descent direction to gradually correct the residuals. By using a preset fixed learning rate, the contribution weight of each newly added decision tree to the final ensemble model prediction result is controlled, so that the weighted predictions of multiple decision trees approximate the actual water level change pattern, thereby achieving prediction and monitoring of water level dynamics under the coupling effect of multiple factors.

[0004] However, when the reservoir basin is in a seasonal climate transition phase (such as the transition from the rainy season to the dry season), weather conditions are often unstable, and noise interference in the water level data may be relatively enhanced. This enhanced noise interference can mask the trend characteristics in the water level data, introducing irregular fluctuations. At this time, the fixed learning rate GBDT model may overfit to these random patterns during training, leading to overfitting. This causes the residual correction direction and weight adjustment to deviate from the true pattern, ultimately resulting in a decrease in the model's generalization ability and prediction accuracy.

[0005] Therefore, how to adaptively adjust the learning rate in the GBDT model to improve its prediction accuracy in complex scenarios such as climate transition periods, and thus achieve dynamic monitoring of reservoir water levels, has become an urgent problem to be solved. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide a method for dynamic monitoring of reservoir water levels based on machine learning, in order to solve the problem of how to adaptively adjust the learning rate in the GBDT model, improve the prediction accuracy of the GBDT model in complex scenarios such as climate transition periods, and thus realize the dynamic monitoring of reservoir water levels.

[0007] This invention provides a method for dynamic monitoring of reservoir water levels based on machine learning, which includes the following steps: Acquire multidimensional monitoring data of the reservoir for each monitoring moment within a preset historical period prior to the current monitoring moment. The multidimensional monitoring data includes upstream water level data, downstream water level data, temperature data, and humidity data. Based on the monitoring time, upstream water level data and downstream water level data within a preset historical period, a three-dimensional spatial coordinate system is constructed. The data points in the three-dimensional spatial coordinate system are fitted to obtain a fitted straight line. Based on the distance difference between each data point in the three-dimensional spatial coordinate system and the fitted straight line, at least one suspicious environmental time interval is selected within the preset historical period. For any suspicious environmental time interval, the multidimensional monitoring data within the suspicious environmental time interval is classified according to the fluctuation characteristics of the multidimensional monitoring data within the suspicious environmental time interval, and the data classification result corresponding to the suspicious environmental time interval is obtained. The data classification results corresponding to each suspicious environmental time interval are obtained. Based on the data classification results corresponding to each suspicious environmental time interval, the preset baseline learning rate in the GBDT model is adjusted to obtain an adaptive learning rate. Based on the adaptive learning rate, an optimized GBDT model is obtained for dynamic monitoring of the reservoir water level at the current monitoring time.

[0008] Preferably, the step of filtering at least one suspicious environmental time interval within a preset historical period based on the distance difference between each data point in the three-dimensional spatial coordinate system and the fitted straight line includes: In the three-dimensional spatial coordinate system, for any data point in the three-dimensional spatial coordinate system, the vertical distance between the data point and the fitted straight line is obtained, and the vertical distance is processed by hyperbolic tangent to obtain the data correlation index of the monitoring time corresponding to the data point. If the data correlation index is greater than or equal to the preset data correlation index threshold, then the monitoring time corresponding to any data point is confirmed as a suspicious monitoring time. Obtain all suspicious monitoring times within a preset historical period, and based on all suspicious monitoring times, filter at least one suspicious environmental time interval within the preset historical period.

[0009] Preferably, the step of filtering at least one suspicious environmental time interval within a preset historical time period based on all suspicious monitoring times includes: All suspicious monitoring moments within a preset historical period are sorted in chronological order to form a set of suspicious monitoring moments. The set of suspicious monitoring moments is then clustered to obtain at least one cluster. For any cluster, obtain the minimum and maximum suspicious monitoring times in the cluster, and construct the suspicious environment time interval corresponding to the cluster with the minimum suspicious monitoring time as the start time and the maximum suspicious monitoring time as the end time.

[0010] Preferably, the step of classifying the multidimensional monitoring data within any suspected environmental time interval based on the fluctuation characteristics of the multidimensional monitoring data within any suspected environmental time interval to obtain the data classification result corresponding to the any suspected environmental time interval includes: Based on the fluctuation characteristics of multidimensional monitoring data within any suspected environmental time interval, the upstream water level fluctuation factor and downstream water level fluctuation factor are obtained for each monitoring moment within any suspected environmental time interval, excluding the last monitoring moment. Based on the upstream water level data fluctuation factor and downstream water level data fluctuation factor at each monitoring time except the last monitoring time within any suspected environmental time interval, obtain the total fluctuation factor of the multidimensional monitoring data at each monitoring time except the last monitoring time within any suspected environmental time interval. For the multidimensional monitoring data at any monitoring time other than the last monitoring time within any suspicious environmental time interval, if the total fluctuation factor of the multidimensional monitoring data is greater than a preset total fluctuation factor threshold, then the multidimensional monitoring data is confirmed to be noise interference data; if the total fluctuation factor of the multidimensional monitoring data is less than or equal to the preset total fluctuation factor threshold, then the multidimensional monitoring data is confirmed to be real data.

[0011] Preferably, the step of obtaining the upstream water level fluctuation factor and the downstream water level fluctuation factor for each monitoring moment (excluding the last monitoring moment) within any suspected environmental time interval based on the fluctuation characteristics of multidimensional monitoring data within that interval includes: Using the monitoring time within any suspected environmental time interval as the horizontal axis and the upstream water level data at each monitoring time as the vertical axis, a two-dimensional coordinate system for upstream water level data is constructed. For any monitoring time other than the last monitoring time, a vector is constructed with the upstream water level data at any monitoring time as the starting point and the upstream water level data at the next monitoring time as the ending point. This vector is denoted as the instantaneous change vector at any monitoring time. The angle between the instantaneous change vector and the horizontal axis of the two-dimensional coordinate system for upstream water level data is obtained and denoted as the angle between the instantaneous change vector of upstream water level at any monitoring time. In the two-dimensional coordinate system of the upstream water level data, obtain the angle between the instantaneous change vectors of the upstream water level at each monitoring moment except the last monitoring moment; Using the monitoring time within any suspicious environmental time interval as the horizontal axis and the temperature data at each monitoring time as the vertical axis, a two-dimensional coordinate system for temperature data is constructed. In the two-dimensional coordinate system for temperature data, the angle between the instantaneous temperature change vectors at each monitoring time except the last monitoring time is obtained. Using the monitoring time within any suspicious environmental time interval as the horizontal axis and the humidity data at each monitoring time as the vertical axis, a two-dimensional coordinate system for humidity data is constructed. In the two-dimensional coordinate system for humidity data, the angle between the instantaneous humidity change vectors at each monitoring time except the last monitoring time is obtained. Based on the angles of the instantaneous changes in upstream water level vectors, the angles of the instantaneous changes in temperature vectors, and the angles of the instantaneous changes in humidity vectors at each monitoring moment, the fluctuation factor of upstream water level data for each monitoring moment except the last monitoring moment is obtained. Using the monitoring time within any suspicious environmental time interval as the horizontal axis and the downstream water level data at each monitoring time as the vertical axis, a two-dimensional coordinate system for downstream water level data is constructed. In the two-dimensional coordinate system for downstream water level data, the angle between the instantaneous change vectors of the next water level at each monitoring time except the last monitoring time is obtained. Based on the angles of the instantaneous changes in downstream water level, temperature, and humidity at each monitoring moment, the fluctuation factor of downstream water level data is obtained for each monitoring moment except the last monitoring moment.

[0012] Preferably, the step of obtaining the upstream water level data fluctuation factor for each monitoring moment except the last monitoring moment, based on the angle of the instantaneous change vector of the upstream water level, the angle of the instantaneous change vector of temperature, and the angle of the instantaneous change vector of humidity at each monitoring moment, includes: For any monitoring time other than the last monitoring time, the average value of the absolute value of the angle between the instantaneous change vectors of the upstream water level at each monitoring time other than the last monitoring time is obtained and denoted as the first mean. The absolute value of the difference between the absolute value of the angle between the instantaneous change vectors of the upstream water level at any monitoring time and the first mean is substituted into the hyperbolic tangent function to obtain the degree of fluctuation of the upstream water level data at any monitoring time. The absolute values ​​of the angles between the instantaneous temperature change vectors and the humidity change vectors at any given monitoring time are obtained and averaged as the second mean. The negative of the absolute value of the difference between the absolute value of the angle between the instantaneous upstream water level change vectors and the second mean at any given monitoring time is substituted into an exponential function with the natural constant as the base to obtain the first multidimensional data correlation degree at any given monitoring time. The upstream water level fluctuation factor at any given monitoring time is obtained by weighted summation of the fluctuation degree of the upstream water level data and the correlation degree of the first multidimensional data.

[0013] Preferably, the step of obtaining the downstream water level data fluctuation factor for each monitoring moment except the last monitoring moment, based on the angle of the instantaneous change vector of the downstream water level, the angle of the instantaneous change vector of temperature, and the angle of the instantaneous change vector of humidity at each monitoring moment, includes: For any monitoring time other than the last monitoring time, the average value of the absolute value of the angle between the instantaneous change vectors of the downstream water level at each monitoring time other than the last monitoring time is obtained and denoted as the third mean. The absolute value of the difference between the absolute value of the angle between the instantaneous change vectors of the downstream water level at any monitoring time and the third mean is substituted into the hyperbolic tangent function to obtain the degree of fluctuation of the downstream water level data at any monitoring time. Substituting the negative of the absolute value of the difference between the instantaneous change vector of the downstream water level at any monitoring time and the second mean into an exponential function with the natural constant as the base, the second multidimensional data correlation degree at any monitoring time is obtained. The downstream water level fluctuation factor at any given monitoring time is obtained by weighted summation of the fluctuation degree of the downstream water level data and the correlation degree of the second multidimensional data.

[0014] Preferably, the step of obtaining the total fluctuation factor of the multidimensional monitoring data for each monitoring time within any suspected environmental time interval, excluding the last monitoring time, based on the upstream water level data fluctuation factor and the downstream water level data fluctuation factor for each monitoring time within any suspected environmental time interval, excluding the last monitoring time, includes: For any monitoring time within any suspected environmental time interval, excluding the last monitoring time, the product between the upstream water level data fluctuation factor and the downstream water level data fluctuation factor at any given monitoring time is obtained to obtain the total fluctuation factor of the multidimensional monitoring data at any given monitoring time.

[0015] Preferably, the step of adjusting the preset baseline learning rate in the GBDT model based on the data classification results corresponding to each suspicious environment time interval to obtain an adaptive learning rate includes: Obtain the proportion of noise interference data in all multidimensional monitoring data within a preset historical time period, calculate the difference between constant 1 and the proportion, obtain the learning rate adjustment coefficient, and obtain the product between the baseline learning rate and the learning rate adjustment coefficient to obtain the adaptive learning rate.

[0016] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows: In this invention, at least one suspicious environmental time interval is selected within a preset historical period to initially locate time periods that may be significantly affected by environmental coupling noise. The data classification results corresponding to the suspicious environmental time intervals are obtained to further identify whether the multidimensional monitoring data within the suspicious environmental time intervals are affected by environmental coupling noise (such as short-term strong convective rainfall) or by actual hydrological fluctuations. Based on this, the preset baseline learning rate in the GBDT model is dynamically adjusted so that the model can adaptively adjust the fitting amplitude under different fluctuation types, improve the prediction accuracy of the model in complex scenarios such as climate transition periods, and achieve highly reliable monitoring and early warning of reservoir water level dynamics. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for dynamic monitoring of reservoir water levels based on machine learning, provided in Embodiment 1 of the present invention. Figure 2 This is a schematic diagram of the two-dimensional coordinate system for upstream water level data provided in Embodiment 1 of the present invention. Detailed Implementation

[0019] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0020] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0021] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0022] See Figure 1 This is a flowchart of a method for dynamic monitoring of reservoir water levels based on machine learning, provided in Embodiment 1 of the present invention. Figure 1 As shown, the method may include: Step S101: Obtain multi-dimensional monitoring data of the reservoir for each monitoring moment within a preset historical period prior to the current monitoring moment. The multi-dimensional monitoring data includes upstream water level data, downstream water level data, temperature data, and humidity data.

[0023] As a core component of water conservancy infrastructure, reservoirs play a crucial role in ensuring their safe and stable operation and improving the intelligence level of water conservancy scheduling through dynamic monitoring of reservoir water levels. Existing methods typically utilize Gradient Boosting Decision Tree (GBDT) models for dynamic monitoring of reservoir water levels. First, initial predictions are obtained using weak learners, and residuals are calculated. Then, environmental factors such as temperature and humidity are considered, and different weights are assigned along the gradient descent direction to gradually correct the residuals. By using a preset fixed learning rate, the contribution weight of each newly added decision tree to the final ensemble model's prediction result is controlled, ensuring that the weighted predictions from multiple decision trees approximate the actual water level changes. This enables the prediction and monitoring of water level dynamics under the coupled effects of multiple factors.

[0024] In this embodiment, a pressure-type water level sensor (such as a piezoresistive sensor) is selected to collect reservoir water level data. Its measurement range covers the historical lowest to highest water level of the reservoir (e.g., 0-100 meters), with an accuracy better than ±0.01 meters. It possesses resistance to wave disturbances, temperature drift compensation, and long-term stability to adapt to dynamic fluctuations in the reservoir. Pressure-type water level sensors are deployed at the control section of the river flowing into the reservoir and near the spillway or outlet in front of the dam to collect upstream and downstream water level data in real time, ensuring that both are at the same elevation datum to accurately reflect the water level gradient between the upstream and downstream of the reservoir. To capture short-term changes during seasonal climate transitions, the sensor's acquisition frequency in this embodiment is once every 12 hours; however, this is not limited and can be set according to the specific implementation scenario.

[0025] Simultaneously, temperature and humidity data are acquired from the standardized public data platform of county-level and above meteorological observation stations within the reservoir's jurisdiction. These data are then timestamped and aligned with upstream and downstream water level data collected by sensors to form multi-dimensional monitoring data for each monitoring moment. This multi-dimensional monitoring data includes upstream water level data, downstream water level data, temperature data, and humidity data. All collected raw data undergoes filtering and noise reduction to remove outliers caused by significant physical interference; therefore, the multi-dimensional monitoring data is the filtered and denoised data.

[0026] Multidimensional monitoring data of the reservoir at each monitoring moment within a preset historical period prior to the current monitoring moment is acquired for dynamic monitoring of the reservoir water level at the current monitoring moment. In order to fully cover the water level and meteorological change characteristics during the climate transition period, while ensuring the sample size requirements and computational efficiency for model training, the preset historical period is set to 100 days in this embodiment. This is not a limitation and can be set according to the specific implementation scenario.

[0027] When a reservoir basin is in a seasonal climate transition phase (such as the transition from the rainy season to the dry season), weather conditions are often unstable, and noise interference in the water level data may be relatively enhanced. This enhanced noise interference can mask the trend characteristics in the water level data, introducing irregular fluctuations. In this case, the fixed-learning-rate GBDT model may overfit to these random patterns during training, leading to overfitting. This causes the residual correction direction and weight adjustment to deviate from the true pattern, ultimately resulting in a decrease in the model's generalization ability and prediction accuracy.

[0028] Therefore, in this embodiment, at least one suspicious environmental time interval is first screened within a preset historical period to initially locate the time interval that may be greatly affected by environmental coupling noise. Then, the data classification results corresponding to the suspicious environmental time interval are obtained to further identify whether the multidimensional monitoring data in the suspicious environmental time interval is affected by environmental coupling noise (such as short-term strong convective rainfall) or by real hydrological fluctuations. Based on this, the preset baseline learning rate in the GBDT model is dynamically adjusted so that the model can adaptively adjust the fitting amplitude under different fluctuation types, improve the prediction accuracy of the model in complex scenarios such as climate transition periods, and achieve highly reliable monitoring and early warning of reservoir water level dynamics.

[0029] Step S102: Based on the monitoring time, upstream water level data and downstream water level data within a preset historical period, a three-dimensional spatial coordinate system is constructed. The data points in the three-dimensional spatial coordinate system are fitted to obtain a fitted straight line. Based on the distance difference between each data point in the three-dimensional spatial coordinate system and the fitted straight line, at least one suspicious environmental time interval is selected within the preset historical period.

[0030] Under stable hydrological conditions, upstream and downstream water level data should maintain a high spatial correlation due to hydraulic connections. However, during climate transition periods, unstable precipitation, localized severe convection, and other environmental noises may have different instantaneous impacts on upstream and downstream water level data, leading to a weakening of the synchronicity of water level changes between the two, resulting in a decrease in spatial correlation.

[0031] Therefore, this embodiment constructs a three-dimensional spatial coordinate system based on the monitoring time, upstream water level data, and downstream water level data within a preset historical period. The X-axis represents the monitoring time, the Y-axis represents the upstream water level data, and the Z-axis represents the downstream water level data. The coordinates of each data point are... This indicates that at monitoring time t, the upstream water level data is... Downstream water level data is Under stable hydrological conditions, data points in a three-dimensional spatial coordinate system should be closely distributed near an ideal spatial straight line representing the coordinated change of water levels upstream and downstream. Therefore, this embodiment uses the least squares method to fit the data points in the three-dimensional spatial coordinate system to obtain a fitted straight line. The least squares method is an existing technology and will not be elaborated here. It is used to represent the baseline of the coordinated change of water levels upstream and downstream within a preset historical period. Based on the distance difference between each data point in the three-dimensional spatial coordinate system and the fitted straight line, at least one suspicious environmental time interval is screened within the preset historical period to initially locate the time period that may be greatly affected by environmental coupling noise.

[0032] The method for filtering at least one suspicious environmental time interval within a preset historical period based on the distance difference between each data point in the three-dimensional spatial coordinate system and the fitted straight line is as follows: In the three-dimensional spatial coordinate system, for any data point in the three-dimensional spatial coordinate system, the perpendicular distance between the data point and the fitted line is obtained. The calculation of the perpendicular distance from a point to a line in the three-dimensional spatial coordinate system is an existing technology and will not be elaborated here. The perpendicular distance is processed by hyperbolic tangent to obtain the data correlation index of the monitoring time corresponding to the data point. The data correlation index quantifies the degree of deviation of the data point from the baseline (i.e. the fitted line) at each monitoring time. The larger the data correlation index, the farther the data point is from the baseline at that monitoring time, the lower the coordination between upstream and downstream water levels, and the weaker the spatial correlation. The smaller the data correlation index, the closer the data point is to the baseline at that monitoring time, the higher the coordination between upstream and downstream water levels, and the stronger the spatial correlation. Therefore, a preset data correlation index threshold is set. If the data correlation index is less than the preset data correlation index threshold, the water level data at the monitoring time corresponding to any data point is determined to be high spatial correlation water level data, which belongs to a stable and coordinated state. If the data correlation index is greater than or equal to the preset data correlation index threshold, the water level data at the monitoring time corresponding to any data point is determined to be low spatial correlation water level data, which may be affected by environmental coupling noise interference. In this case, the monitoring time corresponding to any data point is confirmed as a suspicious monitoring time. In order to improve the sensitivity to low spatial correlation water level data (i.e., water level data that may be affected by noise interference), the preset data correlation index threshold is set to 0.6 in this embodiment. There is no limitation here. In actual applications, the threshold can be finely adjusted according to the hydraulic response characteristics of the specific reservoir. According to the above method for judging suspicious monitoring times, all suspicious monitoring times within a preset historical period are obtained, and all suspicious monitoring times within the preset historical period are sorted in chronological order to form a set of suspicious monitoring times. The set of suspicious monitoring times is then subjected to density-based DBSCAN clustering to obtain at least one cluster. For any cluster, the minimum and maximum suspicious monitoring times in the cluster are obtained. The minimum suspicious monitoring time is used as the start time and the maximum suspicious monitoring time is used as the end time to construct the suspicious environment time interval corresponding to the cluster. Similarly, each cluster obtains a continuous suspicious environment time interval, that is, the start and end time range covered by each cluster constitutes a continuous suspicious environment time interval (the time interval dominated by suspicious environment coupled noise).

[0033] Thus, at least one suspicious environmental time interval has been obtained by filtering within the preset historical time period.

[0034] Step S103: For any suspicious environmental time interval, classify the multidimensional monitoring data within the suspicious environmental time interval according to the fluctuation characteristics of the multidimensional monitoring data within the suspicious environmental time interval, and obtain the data classification result corresponding to the suspicious environmental time interval.

[0035] Environmental coupling noise typically causes brief and violent pulse-like fluctuations in water level data and environmental data, and these fluctuations may have a strong instantaneous coupling relationship between water level and meteorological data (i.e., within a questionable environmental time interval, the pulse-like fluctuations of water level data and environmental data are highly correlated); while systemic hydrological fluctuations usually exhibit more gradual and continuous trend changes, and their coupling strength with short-term meteorological factor fluctuations is relatively weak.

[0036] Therefore, in this embodiment, for any suspicious environmental time interval, the multidimensional monitoring data within the suspicious environmental time interval is classified according to the fluctuation characteristics of the multidimensional monitoring data within the suspicious environmental time interval, and the data classification result corresponding to the suspicious environmental time interval is obtained. This further identifies whether the multidimensional monitoring data within the suspicious environmental time interval is affected by environmental coupling noise (such as short-term strong convective rainfall) or by real hydrological fluctuations, providing a precise basis for subsequent model parameter adjustments.

[0037] The method for classifying the multidimensional monitoring data within any suspicious environmental time interval based on the fluctuation characteristics of the multidimensional monitoring data within that interval, and obtaining the data classification result corresponding to that suspicious environmental time interval, is as follows: (1) Based on the fluctuation characteristics of the multidimensional monitoring data within any suspected environmental time interval, obtain the upstream water level data fluctuation factor and the downstream water level data fluctuation factor for each monitoring moment except the last monitoring moment within any suspected environmental time interval.

[0038] Specifically, a two-dimensional coordinate system for upstream water level data is constructed using the monitoring time within any suspected environmental time interval as the horizontal axis and the upstream water level data at each monitoring time as the vertical axis. For any monitoring time other than the last monitoring time, a vector is constructed with the upstream water level data at that monitoring time as the starting point and the upstream water level data at the next monitoring time as the ending point. This vector is denoted as the instantaneous change vector at that monitoring time. The angle between the instantaneous change vector and the horizontal axis of the two-dimensional coordinate system for upstream water level data is obtained and denoted as the angle between the instantaneous change vector and the upstream water level at that monitoring time. The value range is (-90°, 90°). Constructing the vector and obtaining the angle between the vector and the horizontal axis of the two-dimensional coordinate system for upstream water level data are existing technologies and will not be elaborated here. Two-dimensional coordinate system of upstream water level data, as shown Figure 2 As shown, where, The minimum monitoring time within any of the aforementioned suspicious environmental time intervals. The maximum monitoring time within any suspected environmental time interval is defined as follows: The instantaneous change vector at the minimum monitoring time. The angle between the vectors representing the instantaneous changes in upstream water level at the minimum monitoring time; According to the method for obtaining the angle of the instantaneous change vector of the upstream water level at any of the above monitoring times, the angle of the instantaneous change vector of the upstream water level at each monitoring time except the last monitoring time is obtained in the two-dimensional coordinate system of the upstream water level data. Using the monitoring time within any suspected environmental time interval as the horizontal axis and the temperature data at each monitoring time as the vertical axis, a two-dimensional coordinate system for temperature data is constructed. Following the method for obtaining the angle of the instantaneous change vector of the upstream water level at any monitoring time, the angle of the instantaneous change vector of temperature at each monitoring time except the last monitoring time is obtained in the two-dimensional coordinate system for temperature data. Using the monitoring time within any suspected environmental time interval as the horizontal axis and the humidity data at each monitoring time as the vertical axis, a two-dimensional coordinate system for humidity data is constructed. Following the method for obtaining the angle of the instantaneous change vector of the upstream water level at any monitoring time, the angle of the instantaneous change vector of humidity at each monitoring time except the last monitoring time is obtained in the two-dimensional coordinate system for humidity data. For any monitoring time other than the last monitoring time, the average value of the absolute value of the angle between the instantaneous change vectors of the upstream water level at each monitoring time other than the last monitoring time is obtained and denoted as the first mean. The absolute value of the difference between the absolute value of the angle between the instantaneous change vectors of the upstream water level at any monitoring time and the first mean is substituted into the hyperbolic tangent function to obtain the degree of fluctuation of the upstream water level data at any monitoring time. The absolute values ​​of the angles between the instantaneous temperature change vectors and the humidity change vectors at any given monitoring time are obtained and averaged as the second mean. The negative of the absolute value of the difference between the absolute value of the angle between the instantaneous upstream water level change vectors and the second mean at any given monitoring time is substituted into an exponential function with the natural constant as the base to obtain the first multidimensional data correlation degree at any given monitoring time. The upstream water level fluctuation factor at any given monitoring time is obtained by weighted summation of the fluctuation degree of the upstream water level data and the correlation degree of the first multidimensional data.

[0039] In one embodiment, taking the i-th monitoring time as an example, the formula for calculating the upstream water level fluctuation factor at the i-th monitoring time is:

[0040] in, Let be the fluctuation factor of the upstream water level data at the i-th monitoring time. Let be the angle between the instantaneous change vectors of the upstream water level at the i-th monitoring moment; This is the first mean; Let be the angle between the instantaneous temperature change vectors at the i-th monitoring moment; Let be the angle between the instantaneous humidity change vectors at the i-th monitoring moment; It is the hyperbolic tangent function; It is an exponential function with the natural constant as its base; It is the absolute value symbol; These are the weighting coefficients.

[0041] It should be noted that, The fluctuation level of the upstream water level at the i-th monitoring time. The degree of fluctuation in upstream water level at the i-th monitoring time was quantified relative to the overall average level. The larger the value, the more drastic the upstream water level fluctuation at the i-th monitoring time is compared to the historical average, exhibiting a clear pulse-like characteristic. The larger it is; The smaller the value, the more gradual the upstream water level fluctuation at the i-th monitoring time, with no obvious pulse-like characteristics. The smaller it is; The first multidimensional data correlation degree at the i-th monitoring time quantifies the instantaneous coupling strength between the upstream water level fluctuation and the temperature and humidity fluctuation direction and amplitude at the i-th monitoring time: The larger the value, the more consistent the instantaneous change patterns of water level fluctuations and temperature and humidity fluctuations are at the i-th monitoring time, indicating a strong coupling correlation. The smaller the value, the less consistent the instantaneous fluctuation patterns of the water level fluctuation and the temperature and humidity fluctuation at the i-th monitoring time are, indicating a weak coupling correlation. The weighting coefficient is used to adjust the contribution of pulse intensity and coupling correlation features to the upstream water level data fluctuation factor. To assign higher weight to pulse features and more sensitively capture typical severe pulse signals caused by environmental coupling noise, this embodiment sets... There are no restrictions here; the settings can be adjusted according to the specific implementation scenario. The fluctuation factor of the upstream water level data at the i-th monitoring time integrates pulse intensity and coupling correlation. The larger the value, the higher the probability that the water level fluctuation at the i-th monitoring time has both the characteristics of drastic pulse-like changes and high coupling with meteorological factors, and that it belongs to the false fluctuation dominated by environmental coupling noise. The smaller the value, the higher the probability that the water level fluctuation at the i-th monitoring time is a real fluctuation driven by systematic hydrology.

[0042] According to the method for obtaining the upstream water level data fluctuation factor at the i-th monitoring time, the upstream water level data fluctuation factor is obtained for each monitoring time except the last monitoring time within any suspected environmental time interval.

[0043] Similarly, using the monitoring time within any suspected environmental time interval as the horizontal axis and the downstream water level data at each monitoring time as the vertical axis, a two-dimensional coordinate system for downstream water level data is constructed. Following the method for obtaining the angle of the instantaneous change vector of the upstream water level at any monitoring time, the angle of the next instantaneous change vector of the water level at each monitoring time other than the last monitoring time is obtained in the two-dimensional coordinate system for downstream water level data. For any monitoring time other than the last monitoring time, the average value of the absolute value of the angle between the instantaneous change vectors of the downstream water level at each monitoring time other than the last monitoring time is obtained and denoted as the third mean. The absolute value of the difference between the absolute value of the angle between the instantaneous change vectors of the downstream water level at any monitoring time and the third mean is substituted into the hyperbolic tangent function to obtain the degree of fluctuation of the downstream water level data at any monitoring time. Substituting the negative of the absolute value of the difference between the instantaneous change vector of the downstream water level at any monitoring time and the second mean into an exponential function with the natural constant as the base, the second multidimensional data correlation degree at any monitoring time is obtained. The downstream water level fluctuation factor at any given monitoring time is obtained by weighted summation of the fluctuation degree of the downstream water level data and the correlation degree of the second multidimensional data.

[0044] According to the method for obtaining the downstream water level data fluctuation factor at any given monitoring time, the downstream water level data fluctuation factor is obtained for each monitoring time within any suspected environmental time interval, except for the last monitoring time.

[0045] (2) Based on the upstream water level data fluctuation factor and downstream water level data fluctuation factor of each monitoring time except the last monitoring time within any suspected environmental time interval, obtain the total fluctuation factor of the multidimensional monitoring data of each monitoring time except the last monitoring time within any suspected environmental time interval.

[0046] Specifically, for any monitoring time within any suspected environmental time interval, excluding the last monitoring time, the product between the upstream water level data fluctuation factor and the downstream water level data fluctuation factor at any given monitoring time is obtained to obtain the total fluctuation factor of the multidimensional monitoring data at any given monitoring time.

[0047] In one embodiment, taking the i-th monitoring time as an example, the formula for calculating the total fluctuation factor at the i-th monitoring time is:

[0048] in, The total fluctuation factor at the i-th monitoring time; Let be the fluctuation factor of the upstream water level data at the i-th monitoring time. Let be the fluctuation factor of the downstream water level data at the i-th monitoring time.

[0049] It should be noted that, and The larger the value, the higher the probability that the water level fluctuation at the i-th monitoring time is a spurious fluctuation dominated by environmental coupled noise; and The smaller the value, the higher the probability that the water level fluctuation at the i-th monitoring time is a real fluctuation driven by systematic hydrology.

[0050] (3) Based on the total fluctuation factor of the multidimensional monitoring data of each monitoring time except the last monitoring time within any suspicious environment time interval, classify the multidimensional monitoring data within any suspicious environment time interval to obtain the data classification result corresponding to any suspicious environment time interval.

[0051] Specifically, for the multidimensional monitoring data at any monitoring time other than the last monitoring time within any suspected environmental time interval, if the total fluctuation factor of the multidimensional monitoring data is greater than a preset total fluctuation factor threshold, it indicates that the instantaneous changes in water level data in the multidimensional monitoring data are mainly driven by environmental coupled noise, and the multidimensional monitoring data is confirmed to be noise interference data. If the total fluctuation factor of the multidimensional monitoring data is less than or equal to the preset total fluctuation factor threshold, it indicates that the instantaneous changes in water level data in the multidimensional monitoring data are mainly driven by systematic hydrological processes, and the multidimensional monitoring data is confirmed to be real data. In order to strictly screen out noise pulses with high confidence and avoid misjudging real hydrological fluctuations, the preset total fluctuation factor threshold is set to 0.7 in this embodiment. This is not a limitation and can be set according to the specific implementation scenario.

[0052] Similarly, the classification results of the multi-dimensional monitoring data for each monitoring moment within any suspicious environment time interval, excluding the last monitoring moment, are obtained and used as the data classification results corresponding to any suspicious environment time interval.

[0053] Step S104: Obtain the data classification results corresponding to each suspicious environment time interval. Based on the data classification results corresponding to each suspicious environment time interval, adjust the preset baseline learning rate in the GBDT model to obtain an adaptive learning rate. Based on the adaptive learning rate, obtain an optimized GBDT model for dynamic monitoring of the reservoir water level at the current monitoring time.

[0054] Following the method described above for obtaining data classification results for any suspicious environmental time interval, the data classification results for each suspicious environmental time interval are obtained. Based on these results, the preset baseline learning rate in the GBDT model is adjusted to obtain an adaptive learning rate. This allows the model structure to adapt to different data quality stages, suppressing noise interference while maintaining the ability to capture core patterns, thereby improving overall prediction accuracy and generalization performance. Furthermore, based on the adaptive learning rate, an optimized GBDT model is obtained to dynamically monitor the reservoir water level at the current monitoring time.

[0055] The method for adjusting the preset baseline learning rate in the GBDT model based on the data classification results corresponding to each suspicious environment time interval is as follows: Obtain the proportion of noise interference data in all multidimensional monitoring data within a preset historical time period, calculate the difference between constant 1 and the proportion, obtain the learning rate adjustment coefficient, and obtain the product between the baseline learning rate and the learning rate adjustment coefficient to obtain the adaptive learning rate.

[0056] In one implementation, the adaptive learning rate is calculated using the following formula:

[0057] in, Adaptive learning rate; Use the baseline learning rate; The number of noise interference data within a preset historical period (i.e., the number of multidimensional monitoring data whose total fluctuation factor is greater than the preset total fluctuation factor threshold); F is the total fluctuation factor. The threshold for the total volatility factor is set; N is the number of multidimensional monitoring data within the preset historical period.

[0058] It should be noted that, This sets the percentage of noise interference data within a predefined historical time period in all multidimensional monitoring data. The larger the value, the higher the proportion of environmental coupling noise within the preset historical period. In this case, the baseline learning rate should be reduced so that the model updates with a smaller step size, avoiding the model from overlearning the random patterns introduced by noise, thereby reducing the risk of overfitting, weakening the fitting strength of noise features, and enhancing its robustness. The smaller the value, the lower the proportion of environmental coupling noise. In this case, the model needs to be trained with a learning rate close to the baseline learning rate to ensure that the model can effectively learn the main trends and converge quickly, and learn the main patterns in the data.

[0059] Based on an adaptive learning rate, an optimized GBDT model is obtained. Dynamic monitoring of the reservoir water level at the current monitoring moment is an existing technology, which will be briefly described here: (1) Model prediction and residual calculation: The model first makes an initial prediction on the training data using a simple weak learner (such as based on simple statistical mean), and calculates the residual between the initial prediction value and the true value; (2) Iterative Residuals: In each subsequent iteration, a new weak learner (usually a decision tree) is trained. This weak learner uses the residual sequence from the previous iteration as the fitting object and simultaneously uses current meteorological data (such as temperature and humidity) as key input features to learn, thereby capturing the dynamic influence of environmental factors on the water level fluctuation residuals. The entire process is carried out under the control of an adaptive learning rate, which serves as the update step size. This precisely controls the correction magnitude of the prediction results of each new model to the current cumulative prediction value, improving the model's approximation ability while ensuring its generalization stability.

[0060] (3) Strong learner ensemble prediction: After multiple iterations, the model forms the final prediction through weighted ensemble: the initial prediction value and the prediction output of each subsequent decision tree are weighted according to their respective dynamic learning rates and accumulated round by round. The results of all weak learners are linearly combined to form a powerful ensemble model with better fitting ability, thereby outputting the final and more accurate prediction value of the water level data.

[0061] The optimized GBDT model can fit the normal hydrological change pattern more accurately. The more accurate its prediction, the better the calculated residual sequence can reflect the fluctuations that the model cannot explain. By monitoring whether this residual sequence significantly exceeds the confidence range based on historical normal data statistics, it can be determined with high confidence that it is caused by non-climate abnormal factors (such as engineering hazards or sudden discharge), thereby triggering an early warning and realizing dynamic monitoring of reservoir water levels.

[0062] In summary, in this embodiment of the invention, at least one suspicious environmental time interval is screened within a preset historical period to initially locate time periods that may be significantly affected by environmental coupling noise; the data classification results corresponding to the suspicious environmental time intervals are obtained to further identify whether the multidimensional monitoring data within the suspicious environmental time intervals are affected by environmental coupling noise (such as short-term strong convective rainfall) or by actual hydrological fluctuations, and the preset baseline learning rate in the GBDT model is dynamically adjusted accordingly, so that the model can adaptively adjust the fitting amplitude under different fluctuation types, improve the prediction accuracy of the model in complex scenarios such as climate transition periods, and achieve highly reliable monitoring and early warning of reservoir water level dynamics.

[0063] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for dynamic monitoring of reservoir water levels based on machine learning, characterized in that, The machine learning-based method for dynamic monitoring of reservoir water levels includes: Acquire multidimensional monitoring data of the reservoir for each monitoring moment within a preset historical period prior to the current monitoring moment. The multidimensional monitoring data includes upstream water level data, downstream water level data, temperature data, and humidity data. Based on the monitoring time, upstream water level data and downstream water level data within a preset historical period, a three-dimensional spatial coordinate system is constructed. The data points in the three-dimensional spatial coordinate system are fitted to obtain a fitted straight line. Based on the distance difference between each data point in the three-dimensional spatial coordinate system and the fitted straight line, at least one suspicious environmental time interval is selected within the preset historical period. For any suspicious environmental time interval, the multidimensional monitoring data within the suspicious environmental time interval is classified according to the fluctuation characteristics of the multidimensional monitoring data within the suspicious environmental time interval, and the data classification result corresponding to the suspicious environmental time interval is obtained. The data classification results corresponding to each suspicious environmental time interval are obtained. Based on the data classification results corresponding to each suspicious environmental time interval, the preset baseline learning rate in the GBDT model is adjusted to obtain an adaptive learning rate. Based on the adaptive learning rate, an optimized GBDT model is obtained for dynamic monitoring of the reservoir water level at the current monitoring time.

2. The method for dynamic monitoring of reservoir water levels based on machine learning according to claim 1, characterized in that, The step of filtering at least one suspicious environmental time interval within a preset historical period based on the distance difference between each data point in the three-dimensional spatial coordinate system and the fitted straight line includes: In the three-dimensional spatial coordinate system, for any data point in the three-dimensional spatial coordinate system, the vertical distance between the data point and the fitted straight line is obtained, and the vertical distance is processed by hyperbolic tangent to obtain the data correlation index of the monitoring time corresponding to the data point. If the data correlation index is greater than or equal to the preset data correlation index threshold, then the monitoring time corresponding to any data point is confirmed as a suspicious monitoring time. Obtain all suspicious monitoring times within a preset historical period, and based on all suspicious monitoring times, filter at least one suspicious environmental time interval within the preset historical period.

3. The reservoir water level dynamic monitoring method based on machine learning according to claim 2, characterized in that, The step of filtering at least one suspicious environmental time interval within a preset historical period based on all suspicious monitoring times includes: All suspicious monitoring moments within a preset historical period are sorted in chronological order to form a set of suspicious monitoring moments. The set of suspicious monitoring moments is then clustered to obtain at least one cluster. For any cluster, obtain the minimum and maximum suspicious monitoring times in the cluster, and construct the suspicious environment time interval corresponding to the cluster with the minimum suspicious monitoring time as the start time and the maximum suspicious monitoring time as the end time.

4. The method for dynamic monitoring of reservoir water levels based on machine learning according to claim 1, characterized in that, The step of classifying the multidimensional monitoring data within any suspected environmental time interval based on the fluctuation characteristics of the multidimensional monitoring data within that interval, and obtaining the data classification result corresponding to that suspected environmental time interval, includes: Based on the fluctuation characteristics of multidimensional monitoring data within any suspected environmental time interval, the upstream water level fluctuation factor and downstream water level fluctuation factor are obtained for each monitoring moment within any suspected environmental time interval, excluding the last monitoring moment. Based on the upstream water level data fluctuation factor and downstream water level data fluctuation factor at each monitoring time except the last monitoring time within any suspected environmental time interval, obtain the total fluctuation factor of the multidimensional monitoring data at each monitoring time except the last monitoring time within any suspected environmental time interval. For the multidimensional monitoring data at any monitoring time other than the last monitoring time within any suspicious environmental time interval, if the total fluctuation factor of the multidimensional monitoring data is greater than a preset total fluctuation factor threshold, then the multidimensional monitoring data is confirmed to be noise interference data; if the total fluctuation factor of the multidimensional monitoring data is less than or equal to the preset total fluctuation factor threshold, then the multidimensional monitoring data is confirmed to be real data.

5. The reservoir water level dynamic monitoring method based on machine learning according to claim 4, characterized in that, The step of obtaining the upstream water level fluctuation factor and downstream water level fluctuation factor for each monitoring moment (excluding the last monitoring moment) within any suspected environmental time interval based on the fluctuation characteristics of multidimensional monitoring data within that interval includes: Using the monitoring time within any suspected environmental time interval as the horizontal axis and the upstream water level data at each monitoring time as the vertical axis, a two-dimensional coordinate system for upstream water level data is constructed. For any monitoring time other than the last monitoring time, a vector is constructed with the upstream water level data at any monitoring time as the starting point and the upstream water level data at the next monitoring time as the ending point. This vector is denoted as the instantaneous change vector at any monitoring time. The angle between the instantaneous change vector and the horizontal axis of the two-dimensional coordinate system for upstream water level data is obtained and denoted as the angle between the instantaneous change vector of upstream water level at any monitoring time. In the two-dimensional coordinate system of the upstream water level data, obtain the angle between the instantaneous change vectors of the upstream water level at each monitoring moment except the last monitoring moment; Using the monitoring time within any suspicious environmental time interval as the horizontal axis and the temperature data at each monitoring time as the vertical axis, a two-dimensional coordinate system for temperature data is constructed. In the two-dimensional coordinate system for temperature data, the angle between the instantaneous temperature change vectors at each monitoring time except the last monitoring time is obtained. Using the monitoring time within any suspicious environmental time interval as the horizontal axis and the humidity data at each monitoring time as the vertical axis, a two-dimensional coordinate system for humidity data is constructed. In the two-dimensional coordinate system for humidity data, the angle between the instantaneous humidity change vectors at each monitoring time except the last monitoring time is obtained. Based on the angles of the instantaneous changes in upstream water level vectors, the angles of the instantaneous changes in temperature vectors, and the angles of the instantaneous changes in humidity vectors at each monitoring moment, the fluctuation factor of upstream water level data for each monitoring moment except the last monitoring moment is obtained. Using the monitoring time within any suspicious environmental time interval as the horizontal axis and the downstream water level data at each monitoring time as the vertical axis, a two-dimensional coordinate system for downstream water level data is constructed. In the two-dimensional coordinate system for downstream water level data, the angle between the instantaneous change vectors of the next water level at each monitoring time except the last monitoring time is obtained. Based on the angles of the instantaneous changes in downstream water level, temperature, and humidity at each monitoring moment, the fluctuation factor of downstream water level data is obtained for each monitoring moment except the last monitoring moment.

6. The reservoir water level dynamic monitoring method based on machine learning according to claim 5, characterized in that, The process of obtaining the upstream water level fluctuation factor for each monitoring moment (excluding the last monitoring moment) based on the angles of the instantaneous changes in upstream water level, temperature, and humidity vectors at each monitoring moment includes: For any monitoring time other than the last monitoring time, the average value of the absolute value of the angle between the instantaneous change vectors of the upstream water level at each monitoring time other than the last monitoring time is obtained and denoted as the first mean. The absolute value of the difference between the absolute value of the angle between the instantaneous change vectors of the upstream water level at any monitoring time and the first mean is substituted into the hyperbolic tangent function to obtain the degree of fluctuation of the upstream water level data at any monitoring time. The absolute values ​​of the angles between the instantaneous temperature change vectors and the humidity change vectors at any given monitoring time are obtained and averaged as the second mean. The negative of the absolute value of the difference between the absolute value of the angle between the instantaneous upstream water level change vectors and the second mean at any given monitoring time is substituted into an exponential function with the natural constant as the base to obtain the first multidimensional data correlation degree at any given monitoring time. The upstream water level fluctuation factor at any given monitoring time is obtained by weighted summation of the fluctuation degree of the upstream water level data and the correlation degree of the first multidimensional data.

7. The reservoir water level dynamic monitoring method based on machine learning according to claim 6, characterized in that, The process of obtaining the downstream water level fluctuation factor for each monitoring moment (excluding the last monitoring moment) based on the angles of the instantaneous changes in downstream water level, temperature, and humidity vectors at each monitoring moment includes: For any monitoring time other than the last monitoring time, the average value of the absolute value of the angle between the instantaneous change vectors of the downstream water level at each monitoring time other than the last monitoring time is obtained and denoted as the third mean. The absolute value of the difference between the absolute value of the angle between the instantaneous change vectors of the downstream water level at any monitoring time and the third mean is substituted into the hyperbolic tangent function to obtain the degree of fluctuation of the downstream water level data at any monitoring time. Substituting the negative of the absolute value of the difference between the instantaneous change vector of the downstream water level at any monitoring time and the second mean into an exponential function with the natural constant as the base, the second multidimensional data correlation degree at any monitoring time is obtained. The downstream water level fluctuation factor at any given monitoring time is obtained by weighted summation of the fluctuation degree of the downstream water level data and the correlation degree of the second multidimensional data.

8. The method for dynamic monitoring of reservoir water levels based on machine learning according to claim 1, characterized in that, The step of obtaining the total fluctuation factor of multidimensional monitoring data for each monitoring moment within any suspected environmental time interval (excluding the last monitoring moment) based on the upstream water level data fluctuation factor and the downstream water level data fluctuation factor for each monitoring moment within any suspected environmental time interval (excluding the last monitoring moment) includes: For any monitoring time within any suspected environmental time interval, excluding the last monitoring time, the product between the upstream water level data fluctuation factor and the downstream water level data fluctuation factor at any given monitoring time is obtained to obtain the total fluctuation factor of the multidimensional monitoring data at any given monitoring time.

9. The method for dynamic monitoring of reservoir water levels based on machine learning according to claim 1, characterized in that, The step of adjusting the preset baseline learning rate in the GBDT model based on the data classification results corresponding to each suspicious environment time interval to obtain an adaptive learning rate includes: Obtain the proportion of noise interference data in all multidimensional monitoring data within a preset historical time period, calculate the difference between constant 1 and the proportion, obtain the learning rate adjustment coefficient, and obtain the product between the baseline learning rate and the learning rate adjustment coefficient to obtain the adaptive learning rate.

Images