Distributed reservoir group joint dispatching signal transmission and intelligent decision system

Abstract

The present application relates to the field of hydraulic engineering, particularly to a distributed reservoir group joint scheduling signal transmission and intelligent decision system, comprising a hydrological monitoring layer, a data transmission layer and a scheduling decision layer. The hydrological monitoring layer is composed of multiple hydrological monitoring terminals distributed around the reservoir group, using an ARM Cortex-M4 processor and a ZigBee communication protocol; the data transmission layer uses edge computing nodes and self-organizing networks to realize data preprocessing, encryption and reliable transmission; the scheduling decision layer includes a probability statistics prediction engine and a risk perception decision unit, which realizes the paradigm shift from deterministic prediction to probabilistic prediction through multi-dimensional conditional probability distribution model, uncertainty quantification, multi-time scale prediction integration, probability constraint optimization and stochastic dynamic programming, etc. The system introduces an adaptive learning mechanism to continuously optimize the prediction and decision model, which can reduce the short-term prediction error.

Description

Distributed Reservoir Group Joint Scheduling Signal Transmission and Intelligent Decision-Making System Technical Field

[0001] This invention relates to the field of water conservancy engineering, and in particular to a distributed reservoir group joint scheduling signal transmission and intelligent decision-making system, which is used to realize real-time monitoring, data transmission and intelligent joint scheduling decision-making of reservoir groups. Background Technology

[0002] With the rapid development of water conservancy projects in my country, the joint operation of multiple reservoirs within a river basin is of great significance for flood control and disaster reduction, optimal allocation of water resources, and efficient utilization of hydropower. However, the existing reservoir group operation system faces the following main problems:

[0003] First, traditional reservoir scheduling relies primarily on deterministic prediction models, which cannot effectively address the randomness and uncertainty of hydrological processes. Under complex climatic conditions, prediction errors are significant, affecting the scientific validity and reliability of scheduling decisions.

[0004] Secondly, the existing hydrological monitoring network is not well-deployed, and the reliability of data transmission is low, making it difficult to operate stably under severe weather conditions. This is particularly true in remote mountainous areas where sensor power supply and communication issues are especially prominent.

[0005] Furthermore, the joint scheduling of reservoirs often lacks systematic consideration, and decision-making methods are mostly based on experience-based judgments or simple optimizations, making it difficult to maximize the overall benefits of the watershed.

[0006] In addition, existing dispatching systems generally lack effective mechanisms for quantifying uncertainty and managing risks, which can easily lead to decision-making errors under extreme hydrological conditions, increasing flood control risks and water waste.

[0007] Therefore, there is an urgent need to develop a distributed reservoir group joint scheduling system that can effectively cope with hydrological uncertainties, has reliable signal transmission and intelligent decision-making capabilities. Summary of the Invention

[0008] The purpose of this invention is to provide a distributed reservoir group joint scheduling signal transmission and intelligent decision-making system, which aims to solve the problems of high uncertainty in hydrological forecasting, low reliability of data transmission, and insufficient scientific nature of scheduling decisions in the existing technology.

[0009] This invention proposes a distributed reservoir group joint scheduling signal transmission and intelligent decision-making system, comprising:

[0010] The hydrological monitoring layer is used to collect real-time hydrological and meteorological data around the reservoir group. The hydrological monitoring layer includes multiple hydrological monitoring terminals distributed around the reservoir group. Each hydrological monitoring terminal has built-in multiple types of sensors and processing units.

[0011] The data transmission layer is communicatively connected to the hydrological monitoring layer and is used to receive and transmit hydrological and meteorological data collected by the hydrological monitoring terminal. The data transmission layer includes edge computing nodes and a communication network. The edge computing nodes preprocess and encrypt the received hydrological and meteorological data.

[0012] The scheduling decision layer is communicatively connected to the data transmission layer and is used to make joint scheduling decisions for the reservoir group based on the received hydrological and meteorological data. The scheduling decision layer includes a data storage unit, a probability and statistical prediction engine, and a risk perception decision unit.

[0013] The probabilistic statistical prediction engine is used to construct a multidimensional conditional probability distribution model of hydrological elements based on received hydrological and meteorological data and historical data, and generate hydrological prediction results that include uncertainty quantification.

[0014] The risk perception decision unit is used to convert deterministic scheduling constraints into probabilistic constraints based on the prediction results generated by the multidimensional conditional probability distribution model, perform joint scheduling optimization of the reservoir group under probabilistic constraints, and generate a scheduling decision scheme with risk adaptability.

[0015] Preferably, the hydrological monitoring terminal includes:

[0016] The sensing module includes a temperature sensor, a humidity sensor, a water level sensor, a pressure sensor, a flow sensor, and a rain gauge;

[0017] The processing module, electrically connected to the sensing module, includes an MCU module and a terminal communication module. The MCU module uses an ARM Cortex-M4 processor and is used to sample, clean, and analyze the data collected by the sensing module. The terminal communication module includes a wireless communication unit and an encryption unit. The wireless communication unit uses the ZigBee protocol based on IEEE 802.15.4.

[0018] The power supply module, electrically connected to the processing module, includes a solar panel and a backup power supply, and is used to provide power to the hydrological monitoring terminal.

[0019] Preferably, the edge computing node includes:

[0020] The data storage module is used for temporary storage of received hydrological and meteorological data;

[0021] The data preprocessing module is communicatively connected to the data storage module and is used to perform outlier detection, data cleaning, and compression on the stored hydrological and meteorological data.

[0022] The encryption processing module is communicatively connected to the data preprocessing module and is used to perform the MANOESIS encryption algorithm on the processed hydrological and meteorological data.

[0023] The edge analysis module is communicatively connected to the data storage module and is used to perform preliminary analysis of hydrological and meteorological data to generate a brief hydrological situation report.

[0024] The edge communication module is communicatively connected to the encryption processing module and the edge analysis module, and is used to transmit encrypted data and analysis reports to the scheduling decision layer.

[0025] Preferably, the probability and statistical prediction engine includes:

[0026] The data preprocessing unit is used to perform standardization, outlier identification and processing, time window division, and feature vector construction on the received hydrological and meteorological data.

[0027] The probability distribution modeling unit is communicatively connected to the data preprocessing unit and is used to construct a multidimensional conditional probability distribution model of hydrological elements based on the processed data.

[0028] An uncertainty quantification unit, which is communicatively connected to the probability distribution modeling unit, is used to quantify the uncertainty of the prediction results and generate the prediction confidence interval and error covariance matrix.

[0029] The multi-timescale prediction unit is communicatively connected to the probability distribution modeling unit and the uncertainty quantification unit, and is used to generate prediction results at three time scales: hourly, daily, and weekly. The multi-scale prediction results are integrated through a dynamic weighting mechanism.

[0030] Preferably, the risk perception decision unit includes:

[0031] The probability constraint conversion unit is used to convert deterministic constraints in reservoir scheduling into probabilistic constraints.

[0032] The risk assessment unit is communicatively connected to the probability constraint conversion unit and is used to assess the risk metric of each scheduling scheme based on the uncertainty of the prediction.

[0033] The state space construction unit is used to define the state vector containing key variables such as the water level of each reservoir and the upstream water inflow, and to discretize the continuous state space.

[0034] A transition probability matrix construction unit is communicatively connected to the state space construction unit and is used to construct a state transition probability matrix based on historical data and prediction results.

[0035] The stochastic dynamic programming solution unit is communicatively connected to the probability constraint transformation unit, the risk assessment unit, and the transition probability matrix construction unit, and is used to solve the optimal scheduling strategy based on the stochastic dynamic programming method.

[0036] The scheduling scheme generation unit is communicatively connected to the stochastic dynamic programming solution unit and is used to convert the optimization strategy into specific scheduling instructions and send them to the reservoir control system.

[0037] Preferably, the multi-timescale prediction unit includes:

[0038] Hourly forecast sub-units are used to generate short-term, detailed forecasts for 6-24 hours, focusing on capturing rapid changes in hydrological elements;

[0039] Daily-level forecast sub-units are used to generate 1-7 day medium-term forecasts, balancing rapid changes with trend changes;

[0040] Weekly forecast sub-units are used to generate long-term forecasts of 1-4 weeks, focusing on capturing seasonal changes and long-term trends.

[0041] The scale conversion subunit is communicatively connected to the hourly prediction subunit, the daily prediction subunit, and the weekly prediction subunit, and is used to convert and smoothly transition prediction results at different time scales through upsampling and downsampling methods.

[0042] The dynamic weight allocation subunit is communicatively connected to the hourly, daily, and weekly prediction subunits, and is used to dynamically adjust the weights of the results of each prediction subunit based on historical performance assessments and the current hydrological situation.

[0043] Preferably, the risk assessment unit includes:

[0044] The scenario probability calculation module is used to calculate the probability of occurrence of various possible scenarios based on the predicted probability distribution;

[0045] The consequences severity assessment module is used to assess the severity of constraint violations in various scenarios;

[0046] The risk measurement construction module is communicatively connected to the scenario probability calculation module and the consequence severity assessment module, and is used to combine probability and severity to construct a comprehensive risk measurement.

[0047] The risk tolerance adjustment module is communicatively connected to the risk measurement construction module and is used to dynamically adjust the risk tolerance based on the current water conditions and seasonal characteristics.

[0048] Preferably, the scheduling decision layer further includes an adaptive learning module, which is communicatively connected to the probability statistical prediction engine and the risk perception decision unit. The adaptive learning module is used for:

[0049] Monitor the deviation between the predicted results and the actual observed values;

[0050] Analyze the causes of the deviation;

[0051] Adjust the parameters of the prediction model based on deviation analysis;

[0052] Evaluate the effectiveness of the scheduling strategy;

[0053] Optimize decision model parameters based on execution results;

[0054] Add new experiential knowledge to the system knowledge base.

[0055] Preferably, the communication network adopts a ZigBee self-organizing network, including:

[0056] A network coordinator is used to form and manage a network;

[0057] Routing nodes, which are communicatively connected to the network coordinator, are used to forward data packets and extend network coverage;

[0058] The terminal node is communicatively connected to the routing node and is used to collect and transmit hydrological data;

[0059] The ZigBee self-organizing network supports both single-hop and multi-hop transmission modes, uses the CSMA-CA mechanism for channel access and collision avoidance, and manages wireless data transmission through a superframe structure.

[0060] Preferably, the stochastic dynamic programming solution unit includes:

[0061] The scenario tree construction module is used to generate a multi-branch tree structure representing possible future hydrological scenarios based on the predicted probability distribution.

[0062] The value function solving module is communicatively connected to the scenario tree construction module and is used to iteratively calculate the optimal value function through the Bellman equation.

[0063] The strategy extraction module is communicatively connected to the value function solving module and is used to derive the scheduling strategy from the optimal value function.

[0064] The robust policy evaluation module, which is connected in communication with the policy extraction module, is used to calculate the regret of each policy under different scenarios and select the policy with the minimum maximum regret as the final scheduling policy.

[0065] This invention, by constructing a hierarchical distributed architecture, introducing probabilistic statistical prediction methods and risk perception decision-making mechanisms, has achieved a fundamental shift from a deterministic to a probabilistic paradigm in the joint scheduling of reservoir groups, significantly improving the safety, economy, and ecological benefits of reservoir group operation.

[0066] The beneficial effects of this invention include:

[0067] 1. Improved prediction accuracy and reliability: Through a multidimensional probability distribution prediction model, the system of this invention can accurately characterize the complex correlation between hydrological variables and provide reliable risk assessment through uncertainty quantification. The average error of short-term prediction (within 24 hours) is reduced by 40%, and the coverage of prediction confidence interval is increased to over 95%.

[0068] 2. Enhance the scientific nature and robustness of decision-making: This invention introduces probabilistic constraints and risk assessment mechanisms, incorporates uncertainty into the decision-making process, and seeks the long-term optimal strategy through stochastic dynamic programming, thereby improving the stability of the scheduling scheme under extreme conditions by 70% and avoiding frequent oscillations and adjustments in traditional scheduling.

[0069] 3. Improved system reliability and adaptability: This invention adopts a hierarchical distributed architecture and a self-organizing network, which improves the overall reliability of the system. It can cope with 15% sensor failures and 30% communication interruptions, while still maintaining basic prediction and decision-making capabilities.

[0070] 4. Significant comprehensive benefits: By optimizing the joint scheduling of reservoir groups, the system of this invention significantly improves the comprehensive benefits of flood control and disaster reduction, power generation efficiency and ecological protection. The risk of exceeding the warning water level is reduced by 60%, the annual power generation is increased by 8-15%, and the water supply guarantee rate is increased to 99.5%. Attached Figure Description

[0071] Figure 1 is a schematic diagram of the overall architecture of the distributed reservoir group joint scheduling signal transmission and intelligent decision-making system of the present invention.

[0072] Figure 2 is a schematic diagram of the hydrological monitoring layer of the present invention;

[0073] Figure 3 is a schematic diagram of the data transmission layer of the present invention;

[0074] Figure 4 is a schematic diagram of the scheduling decision layer of the present invention;

[0075] Figure 5 is a schematic diagram of the probability and statistics prediction engine of the present invention;

[0076] Figure 6 is a schematic diagram of the risk perception decision-making unit of the present invention;

[0077] Figure 7 is a schematic diagram of the workflow of the multi-timescale prediction unit of the present invention;

[0078] Figure 8 is a schematic diagram of the workflow of the stochastic dynamic programming solution unit of the present invention;

[0079] Figure 9 is a schematic diagram of the workflow of the adaptive learning module of the present invention;

[0080] Figure 10 is a comparison of the hydrological prediction and scheduling effects of the system of the present invention in a certain watershed. Detailed Implementation

[0081] Please refer to Figures 1-10. The specific embodiments of the present invention will be described in detail below with reference to the figures.

[0082] As shown in Figure 1, the distributed reservoir group joint scheduling signal transmission and intelligent decision-making system of the present invention includes three main parts: hydrological monitoring layer 1, data transmission layer 2, and scheduling decision-making layer 3.

[0083] The hydrological monitoring layer 1 is responsible for collecting real-time hydrological and meteorological data around the reservoir group; the data transmission layer 2 is responsible for receiving and transmitting the data collected by the hydrological monitoring layer 1; and the scheduling decision layer 3 is responsible for making joint scheduling decisions for the reservoir group based on the received hydrological and meteorological data. This layered architecture design makes the system's responsibilities clear and its layers well-defined, facilitating maintenance and expansion.

[0084] As shown in Figure 2, the hydrological monitoring layer 1 includes multiple hydrological monitoring terminals 11 distributed around the reservoir group. Each hydrological monitoring terminal 11 has built-in sensors of various types and processing units for collecting hydrological and meteorological parameters and performing preliminary processing.

[0085] Preferably, the hydrological monitoring terminal 11 includes a sensing module 111, a processing module 112, and a power supply module 113. The sensing module 111 includes a temperature sensor, a humidity sensor, a water level sensor, a pressure sensor, a flow sensor, and a rain gauge. These sensors collect different hydrological and meteorological parameters to form a comprehensive monitoring network.

[0086] The processing module 112 is electrically connected to the sensing module 111 and includes an MCU module 1121 and a terminal communication module 1122. The MCU module 1121 uses an ARM Cortex-M4 processor with a main frequency of up to 168MHz and a floating-point arithmetic unit, suitable for rapid processing of complex hydrological data. The MCU module 1121 is responsible for sampling, cleaning, and analyzing the data collected by the sensing module 111. For example, for water level data, the sampling frequency is set to once per minute. Data that deviates significantly from historical values ​​by more than 50% is marked and temporarily stored for further verification.

[0087] The terminal communication module 1122 includes a wireless communication unit 11221 and an encryption unit 11222. The wireless communication unit 11221 adopts the ZigBee protocol based on IEEE 802.15.4. This protocol operates in the 2.4-2.4835 GHz ISM band, has a variable data rate of 100-300 kbps, and a maximum effective transmission distance of 2 km. The ZigBee protocol supports self-organizing networks and can automatically complete network formation, route discovery, and data forwarding, making it very suitable for hydrological monitoring networks in complex terrain conditions.

[0088] The encryption unit 11222 uses the MANOESIS algorithm for data encryption. This algorithm achieves high-strength encryption through three steps: data scrambling, diffusion, and CRT masking, ensuring the security of data transmission.

[0089] The power supply module 113 is electrically connected to the processing module 112 and includes a solar panel 1131 and a backup power supply 1132. The solar panel 1131 has an area of ​​400 cm², a conversion efficiency of 20%, and can provide 2W of continuous power under standard sunlight conditions. The backup power supply 1132 uses a 24V lithium-ion battery with a capacity of 10000mAh, which can maintain normal system operation for more than 7 days under no-sunlight conditions.

[0090] As shown in Figure 3, the data transmission layer 2 includes edge computing nodes 21 and a communication network 22. The edge computing nodes 21 preprocess and encrypt the received hydrological and meteorological data, while the communication network 22 is responsible for the reliable transmission of the data.

[0091] In one embodiment of the present invention, the edge computing node 21 includes a data storage module 211, a data preprocessing module 212, an encryption processing module 213, an edge analysis module 214, and an edge communication module 215.

[0092] The data storage module 211 is used for temporary storage of received hydrological and meteorological data. Preferably, this module uses 8GB flash memory, supports cyclic writing, and automatically deletes the oldest data when storage space is insufficient, retaining the historical records of the most recent 30 days.

[0093] The data preprocessing module 212 is communicatively connected to the data storage module 211 and is used to perform outlier detection, data cleaning, and compression on the stored hydrological and meteorological data. Outlier detection employs an interquartile range (IQR)-based method, marking data exceeding Q3 + 1.5IQR or falling below Q1 - 1.5IQR (where Q1 is the first quartile, Q3 is the third quartile, and IQR is the IQR = Q3 - Q1). Data cleaning uses local interpolation to correct missing or outlier values. Data compression uses a Huffman coding algorithm, achieving a compression rate typically of 50%-70%, significantly reducing transmission overhead.

[0094] The encryption processing module 213 is communicatively connected to the data preprocessing module 212 and is used to perform the MANOESIS encryption algorithm on the processed hydrological and meteorological data. The encryption process includes:

[0095] (1) Data shuffling: rearranging the positions of the original data sequence to disrupt its original order;

[0096] (2) Data diffusion: Each data bit affects multiple output bits through the XOR operation;

[0097] (3) CRT mask: The Chinese Remainder Theorem is applied to generate a mask to further enhance the encryption strength.

[0098] The edge analysis module 214 is communicatively connected to the data storage module 211 and is used to perform preliminary analysis of hydrological and meteorological data to generate a brief hydrological situation report. This module uses a lightweight time series analysis algorithm to calculate the rate of change and trend of parameters such as water level and flow, and to identify potential abnormal hydrological conditions.

[0099] The edge communication module 215 is communicatively connected to the encryption processing module 213 and the edge analysis module 214, and is used to transmit encrypted data and analysis reports to the scheduling decision layer 3. This module supports multiple communication methods, prioritizing 4G / 5G networks, and can automatically switch to satellite communication in areas with weak signals to ensure the reliability of data transmission.

[0100] Furthermore, as shown in Figure 3, the communication network 22 adopts a ZigBee self-organizing network, including a network coordinator 221, routing nodes 222, and terminal nodes 223. The network coordinator 221 is responsible for forming and managing the network, allocating network addresses and channel resources; the routing nodes 222 are communicatively connected to the network coordinator 221 and are responsible for forwarding data packets and extending network coverage; the terminal nodes 223 are communicatively connected to the routing nodes 222 and are responsible for collecting and transmitting hydrological data.

[0101] ZigBee self-organizing networks support both single-hop and multi-hop transmission modes. When the transmission distance is short (not exceeding the direct communication range), single-hop transmission is used. When the distance is long and relaying is required, multi-hop transmission is used to forward data through routing node 222. The network uses CSMA-CA (Carrier Sense Multiple Access / Collision Avoidance) mechanism for channel access and collision avoidance. The workflow is as follows: a node first listens to the channel status; if the channel is idle, it sends data; otherwise, it enters a random backoff state, waiting a random period of time before trying again.

[0102] The communication network 22 also employs a superframe structure to manage wireless data transmission. A typical superframe includes: a coordinator signal reception time slot, multiple CSMA channel access time slots, a data channel time slot, and a coordinator transmission time slot. This structure balances real-time performance and network efficiency, making it suitable for applications with high timeliness requirements, such as hydrological monitoring.

[0103] As shown in Figure 4, the scheduling decision layer 3 includes a data storage unit 31, a probabilistic statistical prediction engine 32, and a risk perception decision unit 33. The data storage unit 31 is responsible for storing the received hydrological and meteorological data and historical data; the probabilistic statistical prediction engine 32 is responsible for building a prediction model based on these data; and the risk perception decision unit 33 is responsible for generating scheduling decision schemes based on the prediction results.

[0104] In a preferred embodiment, as shown in FIG5, the probability and statistics prediction engine 32 includes a data preprocessing unit 321, a probability distribution modeling unit 322, an uncertainty quantification unit 323, and a multi-timescale prediction unit 324.

[0105] Data preprocessing unit 321 is responsible for standardizing the received hydrological and meteorological data, identifying and handling outliers, dividing time windows, and constructing feature vectors. Standardization is performed using the Z-score method.

[0106] ,

[0107] Where: z represents the standardized data, dimensionless; x represents the original data, with units determined by specific parameters (e.g., water level in meters, flow rate in m³ / s); μ represents the mean of historical data, with the same units as x; σ represents the standard deviation of historical data, with the same units as x. This standardization method transforms hydrological parameters of different dimensions to the same scale, facilitating subsequent modeling.

[0108] Feature vector construction involves converting raw data into feature vectors that reflect hydrological characteristics, including time features (such as hours, dates, and seasons), trend features (such as the rate of change over the past 24 hours), and periodic features (such as daily cycles and weekly cycles).

[0109] The probability distribution modeling unit 322 is communicatively connected to the data preprocessing unit 321, and is used to construct a multidimensional conditional probability distribution model of hydrological elements based on the processed data. This unit uses a nonparametric kernel density estimation method to construct a multidimensional conditional probability distribution including rainfall, surface runoff, water level, and flow rate.

[0110] ,

[0111] in: Let X be the conditional probability density function, representing the conditional probability density of the feature vector X given historical observation conditions Y, with units of probability density. is a d-dimensional hydrological feature vector, whose components may have different units (such as water level, flow rate, etc.); Y is the historical observation dataset; n is the number of samples, which is dimensionless. H is the kernel function (usually a Gaussian kernel or an Epanechnikov kernel), and the choice of kernel function is based on the characteristics of the data distribution; H is the bandwidth matrix, which controls the smoothness of the kernel function, and the unit is determined by the unit combination of the eigenvector components. The weights are based on time distance, are dimensionless, and satisfy... .

[0112] Bandwidth matrix The smoothing parameter of the kernel function is a key parameter affecting the accuracy of kernel density estimation. This system adopts an adaptive bandwidth method, which dynamically adjusts the smoothing parameter of the kernel function according to the data density region. A larger bandwidth is used in sparse regions and a smaller bandwidth is used in dense regions. The calculation formula is as follows:

[0113] ,

[0114] in: For sample points The local bandwidth matrix at the location, with units of and . same; The basic bandwidth matrix (determined through cross-validation) has units determined by the unit combination of the eigenvector components. This is the probability density estimate at the sample point, expressed in probability density units. To adjust the parameter, a value of 0.2-0.5 is typically used; it is dimensionless. Larger values... The value will reduce bandwidth more in high-density areas, enhancing local details.

[0115] Weighting function Designed as an exponential function based on time decay:

[0116] ,

[0117] in: For current time and sample The difference in observation time, in days; This is the attenuation coefficient, in days. The value is typically between 0.01 and 0.05; the denominator ensures that the total weights are equal to 1. This weighting design gives higher weights to recent observation data and exponentially decreases the weights of longer-term data, effectively solving the problem of non-stationary characteristics of hydrological series.

[0118] Uncertainty quantification unit 323 is communicatively connected to probability distribution modeling unit 322, and is used to quantify the uncertainty of the prediction results, generating prediction confidence intervals and error covariance matrices. This unit adopts a Bayesian statistical framework to quantify parameter uncertainty by calculating the posterior distribution of the prediction parameters.

[0119] ,

[0120] in: For parameters The posterior distribution of , in units of probability density; Let be the likelihood function, representing the likelihood given parameters. The observed data The probability, expressed in units of probability; For parameters The prior distribution of , in units of probability density; This indicates a direct proportional relationship. Posterior distribution. Reflects on the observation data Under the condition, for parameters Uncertainty perception.

[0121] Construct predicted values ​​based on the posterior distribution. Confidence interval:

[0122] ,

[0123] in: Represents the p-quantile of the predicted distribution, with the same units as the predicted variable; The significance level is typically set to 0.05, representing a 95% confidence level. For example, a 95% confidence interval is... This means that there is a 95% probability that the true value will fall within this range.

[0124] The error covariance matrix is ​​used to describe the correlation between prediction errors of different hydrological variables:

[0125] ,

[0126] in: for The covariance matrix of dimension , The number of hydrological variables; Let be the prediction variance of the i-th variable, expressed as the square of the units of that variable; Let be the covariance of the prediction errors for the i-th and j-th variables, expressed as the product of the units of the i-th and j-th variables. For example, if the i-th variable is water level (m) and the j-th variable is flow rate (m³ / s), then... The unit is m·m³ / s.

[0127] The multi-timescale prediction unit 324 is communicatively connected to the probability distribution modeling unit 322 and the uncertainty quantification unit 323. It is used to generate prediction results at three time scales: hourly, daily, and weekly, and integrates the multi-scale prediction results through a dynamic weighting mechanism. As shown in Figure 7, this unit includes an hourly prediction subunit 3241, a daily prediction subunit 3242, a weekly prediction subunit 3243, a scale conversion subunit 3244, and a dynamic weight allocation subunit 3245.

[0128] Hourly forecast subunit 3241 generates short-term, detailed forecasts for 6-24 hours, using high-frequency data with a sampling interval of 15 minutes, focusing on capturing rapid changes in hydrological elements, such as rapid water level rises caused by sudden rainfall. Daily forecast subunit 3242 generates medium-term forecasts for 1-7 days, with a sampling interval of 3-6 hours, balancing rapid changes with trend changes, suitable for guiding daily dispatch decisions. Weekly forecast subunit 3243 generates long-term forecasts for 1-4 weeks, with a sampling interval of 1 day, focusing on capturing seasonal changes and long-term trends, providing a basis for strategic dispatch.

[0129] The scale transformation subunit 3244 is communicatively connected to the three prediction subunits mentioned above, and achieves the transformation and smooth transition of prediction results at different time scales through upsampling and downsampling methods. Upsampling uses cubic spline interpolation to convert low-frequency prediction results into high-frequency time series; downsampling uses a weighted average method to convert high-frequency prediction results into low-frequency time series. Boundary constraints ensure a smooth transition of prediction results at different time scales at the boundaries, avoiding prediction faults.

[0130] The dynamic weight allocation subunit 3245 communicates with the three prediction subunits mentioned above, and dynamically adjusts the weights of the prediction subunit results based on historical performance assessments and the current hydrological situation. The weight calculation formula is as follows:

[0131] ,

[0132] in: For the first The weights of each prediction subunit are dimensionless and satisfy the following conditions: ; This represents the historical root mean square error of this sub-unit under the current hydrological conditions, with the same units as the predicted variables. This is a sensitivity parameter, with units equal to the reciprocal of the units of the predictor variable, typically ranging from 0.5 to 2. The larger the value, the more sensitive the weight is to error differences.

[0133] The formula for calculating the integrated prediction results is as follows:

[0134] ,

[0135] in: To integrate the prediction results, the units are the same as those of the predictor variables; For the first The prediction results of each prediction sub-unit are in the same unit as the prediction variable; For the first The weights of each prediction subunit are dimensionless.

[0136] As shown in Figure 6, the risk perception and decision-making unit 33 includes a probability constraint transformation unit 331, a risk assessment unit 332, a state space construction unit 333, a transition probability matrix construction unit 334, a stochastic dynamic programming solution unit 335, and a scheduling scheme generation unit 336.

[0137] The probability constraint conversion unit 331 is used to convert deterministic constraints in reservoir operation into probabilistic constraints. Deterministic constraints in traditional reservoir operation, such as "the water level must not exceed the flood control limit level," are converted into probabilistic form in this system: "the probability that the water level exceeds the flood control limit level must not be greater than..." The expression is:

[0138] ,

[0139] in: Indicates the reservoir water level Exceeding the flood control limit water level The probability is dimensionless; This refers to the reservoir water level, in meters (m). The water level is the flood control limit, and the unit is meters (m). This is the risk tolerance level, dimensionless, and is usually set to 0.01-0.05 (depending on the importance of the constraint).

[0140] The risk assessment unit 332 is communicatively connected to the probability constraint conversion unit 331 and is used to assess the risk metric of each scheduling scheme based on the uncertainty of prediction. The risk assessment unit 332 includes a scenario probability calculation module 3321, a consequence severity assessment module 3322, a risk metric construction module 3323, and a risk tolerance adjustment module 3324.

[0141] The scenario probability calculation module 3321 calculates the probability of occurrence of various possible scenarios based on the predicted probability distribution. For a discrete scenario set... ,scene The probability is:

[0142] ,

[0143] in: For the context The probability of occurrence is dimensionless. Let be the probability density function for prediction, with units equal to the reciprocal of the units of the predictor variables; the integration range is the scenario. The range of values ​​for the corresponding predictor variables. For example, for water level prediction, if the scenario... Defined as "water level between 145m and 146m", then ,in Let be the probability density function of the water level, in units of m. .

[0144] The consequences severity assessment module 3322 assesses the severity of constraint violations under each scenario. The severity assessment function is:

[0145] ,

[0146] in: For the context The severity score is dimensionless. For the context Next The degree of violation of a constraint depends on the type of constraint. The weights are dimensionless to constrain the weights. To constrain the quantity; This indicates that only the case of constraint violation is considered; if the constraint is satisfied, the value is 0. For example, for water level constraints, This indicates the extent to which the water level exceeds the limit, expressed in meters (m).

[0147] The risk measurement construction module 3323 communicates with the scenario probability calculation module 3321 and the consequence severity assessment module 3322 to combine probability and severity to construct a comprehensive risk measurement.

[0148] ,

[0149] in: For comprehensive risk measurement, dimensionless; For the context The probability of occurrence is dimensionless. For the context Severity score, dimensionless; This represents the total number of scenarios. This risk metric indicates the expected severity of constraint violations.

[0150] The risk tolerance adjustment module 3324 is communicatively connected to the risk measurement construction module 3323, and dynamically adjusts the risk tolerance based on the current water conditions and seasonal characteristics.

[0151] ,

[0152] in: This is the adjusted risk tolerance, dimensionless. The baseline risk tolerance is dimensionless (typically 0.05). The adjustment factor is dimensionless (usually 0.8). This is a flood season indicator, with a value of 1 during the flood season and 0 during the non-flood season, and it is dimensionless. This means that during the flood season, the risk tolerance will decrease to 20% of the baseline value (i.e., 0.01), reflecting the stricter safety requirements during the flood season.

[0153] State space construction unit 333 is used to define state vectors containing key variables such as water levels in each reservoir and upstream inflow, and to discretize the continuous state space. The state vector is defined as follows:

[0154] ,

[0155] in: For a moment The state vector contains Each component; For the first The reservoir is at all times Water level, in meters (m); For the first The reservoir is at all times Inbound flow rate, in units of ; Number of reservoirs; superscript This represents the transpose of a vector.

[0156] The state space discretization employs a non-uniform grid method, using a denser grid in critical water level ranges (such as near the flood control limit) to improve decision-making accuracy. For example, the grid interval can be set to 0.1m near the flood control limit (within ±0.5m range), while it can be set to 0.5m in other areas.

[0157] The transition probability matrix construction unit 334 is communicatively connected to the state space construction unit 333, and is used to construct the state transition probability matrix based on historical data and prediction results.

[0158] ,

[0159] in: Indicates the state Take action below Afterwards, the system transitions to state. The probability of is dimensionless. and They are time points and State vector; For a moment The action vector. This indicates the outflow rate of each reservoir, in units of... .

[0160] The transition probability matrix is ​​constructed by combining statistical methods based on historical data with prediction results based on physical models, taking into account both historical patterns and current hydrological characteristics.

[0161] The stochastic dynamic programming solution unit 335 is communicatively connected to the probability constraint transformation unit 331, the risk assessment unit 332, and the transition probability matrix construction unit 334, and is used to solve the optimal scheduling strategy based on the stochastic dynamic programming method. The stochastic dynamic programming solution unit 335 includes a scenario tree construction module 3351, a value function solution module 3352, a policy extraction module 3353, and a robust policy evaluation module 3354.

[0162] The scenario tree construction module 3351 is used to generate a multi-branch tree structure representing possible future hydrological scenarios based on the predicted probability distribution. Each node in the tree represents the state at a given time point, and the branches represent state transitions, with each branch having a corresponding probability weight. The structural parameters of the scenario tree are typically set as follows: a time span of 7 days, 5 branches for the first 3 days, and 3 branches for the last 4 days, forming a structure with... A scenario tree with leaf nodes.

[0163] The value function solving module 3352 communicates with the scenario tree construction module 3351, and calculates the optimal value function iteratively through the Bellman equation:

[0164] ,

[0165] in: For state At any moment Value function, unit and reward function The same, usually in terms of economic benefits (yuan); In the state Take action The immediate rewards (such as the weighted sum of power generation benefits, flood control benefits, etc.) are expressed in yuan. The discount factor is dimensionless and typically set to 0.95-0.99, representing the importance of future rewards relative to current rewards; the summation term represents the summation over all possible next states. Expectations Let be the state transition probability.

[0166] The strategy extraction module 3353 is communicatively connected to the value function solving module 3352, and derives the scheduling strategy from the optimal value function.

[0167] ,

[0168] in: Indicates the state The optimal action, i.e. the optimal outbound flow vector, is expressed in units of... ; This represents the action that maximizes the expression within the parentheses. .

[0169] The robust policy evaluation module 3354 is communicatively connected to the policy extraction module 3353. It calculates the regret of each policy under different scenarios and selects the policy with the minimum and maximum regret as the final scheduling policy. Regret is defined as:

[0170] ,

[0171] in: For strategy In the context The degree of regret is expressed in units similar to that of the value function, typically in units of 1 / 2. For the context The optimal value is given below, in yuan. For strategy In the context The value is given below, in yuan. The minimum-maximum regret strategy is:

[0172] ,

[0173] in: The strategy is a minimum-maximum regret strategy; This represents the strategy that minimizes the following expression. ; In all scenarios The maximum value among them. This strategy selects the strategy with the least regret in the worst case, reflecting robustness considerations.

[0174] The scheduling scheme generation unit 336 is communicatively connected to the stochastic dynamic programming solution unit 335, and is used to convert the optimization strategy into specific scheduling instructions and send them to the reservoir control system. The scheduling instructions include the target outflow, power generation plan, and water level control range for each reservoir.

[0175] Furthermore, as shown in Figure 4, the scheduling decision layer 3 also includes an adaptive learning module 34, which is communicatively connected to the probability and statistics prediction engine 32 and the risk perception decision unit 33. The adaptive learning module 34 is responsible for monitoring the deviation between the prediction results and the actual observations, analyzing the causes of the deviation, adjusting the prediction model parameters based on the deviation analysis, evaluating the execution effect of the scheduling strategy, optimizing the decision model parameters based on the execution effect, and adding new experiential knowledge to the system knowledge base.

[0176] The adaptive learning module 34 employs an online learning method to continuously update and optimize system parameters. For example, recursive least squares (RLS) is used to update the prediction model parameters:

[0177] ,

[0178] ,

[0179] ,

[0180] in: For a moment The model parameter vector, the unit of which depends on the specific parameter; For a moment The model parameter vector unit and same; This is the gain vector, with units equal to the model parameter units divided by the predictor variable units. For a moment The actual observed values ​​are in the same units as the predictor variables; For a moment The predicted values ​​are in the same units as the predicted variables; For a moment The input vector, the unit of which depends on the specific input variable; For a moment The covariance matrix is ​​expressed as the unit square of the model parameters divided by the unit square of the input variables; It is the identity matrix; superscript Represents the transpose of a vector or matrix.

[0181] Adaptive learning is a key mechanism for continuous system optimization and self-improvement, enabling the system to continuously improve prediction accuracy and decision quality as operational data accumulates.

[0182] Taking a group of three cascade reservoirs in a certain river basin as an example, the system of this invention is applied for joint scheduling. The basin covers an area of ​​approximately 10,000 square kilometers, with an average annual rainfall of 1,200 millimeters and an annual runoff of approximately 4 billion cubic meters. The total storage capacity of the three reservoirs is 1.5 billion cubic meters, and the installed capacity is 350 megawatts.

[0183] The system deploys 50 hydrological monitoring points in the basin, covering major tributaries and control sections. The hydrological monitoring terminals utilize the configuration of this invention; sensors collect data every 15 minutes, which is then processed by the MCU and transmitted via a ZigBee network to the nearest edge computing node. The edge computing node performs data preprocessing and encryption before transmitting the data to the basin dispatch center via a 4G network.

[0184] The dispatch center's probabilistic statistical prediction engine constructs a multidimensional hydrological probability model of the watershed based on real-time data and historical data (approximately 10 years). The system simultaneously maintains predictions at three time scales: hourly (24 hours), daily (7 days), and weekly (4 weeks), and integrates them through a dynamic weighting mechanism.

[0185] Based on the prediction results, the risk perception and decision-making unit constructed a transition probability matrix containing the states of the three reservoirs and solved the optimal scheduling strategy through stochastic dynamic programming. The system sets the risk tolerance to 0.01 during the flood season and 0.05 during the non-flood season, reflecting the principle of "safety first during the flood season, and benefits considered during the non-flood season".

[0186] In actual operation, the system demonstrated excellent performance: the average relative error of flood peak prediction was reduced to below 15%, about 40% lower than that of traditional deterministic models; the joint scheduling scheme increased the utilization rate of flood control reservoir capacity by 25%, annual power generation by 12%, and water supply security rate to 99.5%. In particular, during a sudden heavy rainfall event, the system provided an early warning 48 hours in advance and adjusted the reservoir operation mode, effectively reducing the flood peak flow and preventing flooding in downstream towns.

[0187] The adaptive learning module continuously monitors the system's performance and optimizes the prediction model and decision-making strategies. For example, the system found that the prediction error was large under certain rainfall patterns. By analyzing historical data, it adjusted the model parameters under the corresponding conditions, improving the prediction accuracy by 30% in such cases.

[0188] The distributed reservoir group joint scheduling signal transmission and intelligent decision-making system of this invention achieves real-time monitoring, reliable communication, and intelligent decision-making for reservoir groups through the organic integration of hydrological monitoring, data transmission, and scheduling decision-making layers. The core innovation of the system lies in the introduction of probabilistic statistical prediction and risk perception decision-making mechanisms, modeling and managing uncertainty as an inherent characteristic of the hydrological system, fundamentally improving prediction accuracy and decision-making quality. Practice has proven that this system has significant advantages in flood control and disaster reduction, optimal water resource allocation, improved power generation efficiency, and ecological environmental protection, providing strong technical support for the scientific scheduling of reservoir groups and the sustainable utilization of water resources.

[0189] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A distributed reservoir group joint scheduling signal transmission and intelligent decision-making system, characterized in that, include: The hydrological monitoring layer is used to collect real-time hydrological and meteorological data around the reservoir group. The hydrological monitoring layer includes multiple hydrological monitoring terminals distributed around the reservoir group. Each hydrological monitoring terminal has built-in multiple types of sensors and processing units. The data transmission layer, which is communicatively connected to the hydrological monitoring layer, is used to receive and transmit hydrological and meteorological data collected by the hydrological monitoring terminal. The data transmission layer includes edge computing nodes and a communication network. The edge computing nodes preprocess and encrypt the received hydrological and meteorological data. The scheduling decision layer, which is communicatively connected to the data transmission layer, is used to make joint scheduling decisions for the reservoir group based on the received hydrological and meteorological data. The scheduling decision layer includes a data storage unit, a probability statistical prediction engine, and a risk perception decision unit. The probabilistic statistical prediction engine includes: a data preprocessing unit for standardizing received hydrological and meteorological data, identifying and processing outliers, dividing time windows, and constructing feature vectors; a probability distribution modeling unit, communicatively connected to the data preprocessing unit, for constructing a multidimensional conditional probability distribution model containing hydrological elements such as rainfall, surface runoff, water level, and flow rate based on the processed data using a nonparametric kernel density estimation method; an uncertainty quantification unit, communicatively connected to the probability distribution modeling unit, for quantifying parameter uncertainty by calculating the posterior distribution of prediction parameters using a Bayesian statistical framework, and generating prediction confidence intervals and error covariance matrices; and a multi-timescale prediction unit, communicatively connected to the probability distribution modeling unit and the uncertainty quantification unit, for generating prediction results at hourly, daily, and weekly timescales, and integrating the multi-scale prediction results through a dynamic weighting mechanism; the risk perception decision-making unit includes: a probability constraint conversion unit for converting deterministic constraints in reservoir scheduling into probabilistic constraints; and a risk assessment unit, communicatively connected to the probability constraint conversion unit. The system comprises the following components: a meta-communication connection for assessing the risk of each scheduling scheme based on the uncertainty of prediction; a state space construction unit for defining state vectors containing key variables such as reservoir water levels and upstream inflow, and discretizing the continuous state space; a transition probability matrix construction unit, communicating with the state space construction unit, for constructing a state transition probability matrix based on historical data and prediction results; a stochastic dynamic programming solution unit, communicating with the probability constraint transformation unit, the risk assessment unit, and the transition probability matrix construction unit, for solving the optimal scheduling strategy based on the stochastic dynamic programming method; the stochastic dynamic programming solution unit includes a scenario tree construction module, a value function solution module, a strategy extraction module, and a robust strategy evaluation module; the value function solution module iteratively calculates the optimal value function using the Bellman equation; the robust strategy evaluation module calculates the regret of each strategy under different scenarios and selects the strategy with the minimum maximum regret as the final scheduling strategy; and a scheduling scheme generation unit, communicating with the stochastic dynamic programming solution unit, for converting the optimized strategy into specific scheduling instructions and sending them to the reservoir control system.

2. The distributed reservoir group joint scheduling signal transmission and intelligent decision-making system according to claim 1, characterized in that, The hydrological monitoring terminal includes: a sensing module, comprising a temperature sensor, a humidity sensor, a water level sensor, a pressure sensor, a flow sensor, and a rain gauge; a processing module, electrically connected to the sensing module, comprising an MCU module and a terminal communication module, wherein the MCU module uses an ARM Cortex-M4 processor for sampling, cleaning, and analyzing the data collected by the sensing module, and the terminal communication module includes a wireless communication unit and an encryption unit, wherein the wireless communication unit uses the ZigBee protocol based on IEEE 802.15.4; and a power supply module, electrically connected to the processing module, comprising a solar panel and a backup power supply for providing power to the hydrological monitoring terminal.

3. The distributed reservoir group joint scheduling signal transmission and intelligent decision-making system according to claim 1, characterized in that, The edge computing node includes: a data storage module for temporarily storing received hydrological and meteorological data; a data preprocessing module, communicatively connected to the data storage module, for performing outlier detection, data cleaning, and compression on the stored hydrological and meteorological data; an encryption processing module, communicatively connected to the data preprocessing module, for executing the MANOESIS encryption algorithm on the processed hydrological and meteorological data; an edge analysis module, communicatively connected to the data storage module, for performing preliminary analysis on the hydrological and meteorological data and generating a brief hydrological situation report; and an edge communication module, communicatively connected to the encryption processing module and the edge analysis module, for transmitting the encrypted data and analysis report to the scheduling decision layer.

4. The distributed reservoir group joint scheduling signal transmission and intelligent decision-making system according to claim 1, characterized in that, The multi-timescale prediction unit includes: an hourly prediction subunit for generating short-term, detailed predictions of 6-24 hours, focusing on capturing rapid changes in hydrological elements; a daily prediction subunit for generating medium-term predictions of 1-7 days, balancing rapid changes with trend changes; a weekly prediction subunit for generating long-term predictions of 1-4 weeks, focusing on capturing seasonal changes and long-term trends; a scale conversion subunit, communicatively connected to the hourly, daily, and weekly prediction subunits, for converting and smoothly transitioning prediction results at different timescales through upsampling and downsampling methods; and a dynamic weight allocation subunit, communicatively connected to the hourly, daily, and weekly prediction subunits, for dynamically adjusting the weights of the results from each prediction subunit based on historical performance assessments and the current hydrological situation.

5. The distributed reservoir group joint scheduling signal transmission and intelligent decision-making system according to claim 1, characterized in that, The risk assessment unit includes: a scenario probability calculation module, used to calculate the probability of occurrence of various possible scenarios based on the predicted probability distribution; a consequence severity assessment module, used to assess the severity of constraint violations under each scenario; a risk metric construction module, communicatively connected to the scenario probability calculation module and the consequence severity assessment module, used to combine probability and severity to construct a comprehensive risk metric; and a risk tolerance adjustment module, communicatively connected to the risk metric construction module, used to dynamically adjust the risk tolerance according to the current water conditions and seasonal characteristics.

6. The distributed reservoir group joint scheduling signal transmission and intelligent decision-making system according to claim 1, characterized in that, The scheduling decision layer also includes an adaptive learning module, which is communicatively connected to the probability and statistics prediction engine and the risk perception decision unit. The adaptive learning module is used to: monitor the deviation between the prediction results and the actual observations; analyze the causes of the deviation; adjust the prediction model parameters based on the deviation analysis; evaluate the execution effect of the scheduling strategy; optimize the decision model parameters based on the execution effect; and add new experiential knowledge to the system knowledge base.

7. The distributed reservoir group joint scheduling signal transmission and intelligent decision-making system according to claim 1, characterized in that, The communication network adopts a ZigBee self-organizing network, including: a network coordinator for forming and managing the network; routing nodes, which are communicatively connected to the network coordinator for forwarding data packets and extending network coverage; and terminal nodes, which are communicatively connected to the routing nodes for collecting and transmitting hydrological data. The ZigBee self-organizing network supports both single-hop and multi-hop transmission modes, uses the CSMA-CA mechanism for channel access and collision avoidance, and manages wireless data transmission through a superframe structure.

Images