A multi-objective algorithm-based intelligent water supply scheduling control method and device

Abstract

This invention discloses an intelligent water supply scheduling and control method and device based on a multi-objective algorithm. The method involves real-time acquisition of operational status data from the Huqiu water supply network; combining historical water usage data to predict water demand at multiple time scales, generating long-term, medium-term, and short-term water demand prediction sequences; constructing and solving a Stackelberg dynamic game model based on the long-term water demand prediction sequence to obtain a long-term reservoir storage and release operation strategy; constructing and solving a Stackelberg dynamic game model based on the medium-term water demand prediction sequence to obtain a multi-source joint scheduling scheme; generating a dynamic adjustment strategy based on the short-term water demand prediction sequence through an improved multi-agent deep deterministic strategy gradient algorithm; and generating optimized scheduling instructions based on the dynamic adjustment strategy and issuing them to the execution equipment to complete the overall intelligent scheduling and control of the water supply system. This invention solves the problems of insufficient handling of water demand uncertainty, poor multi-source coordination, and weak short-term control stability in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of intelligent water management technology, specifically to an intelligent water supply scheduling and control method and device based on a multi-objective algorithm. Background Technology

[0002] Water supply systems are a crucial infrastructure in modern cities, and their operational efficiency impacts the quality of life for urban residents and economic development. With accelerating urbanization and the emergence of water scarcity, optimizing the scheduling of water supply systems has become a key concern for both academia and engineering.

[0003] Traditional water supply scheduling methods are mainly based on rule-based control and static optimization, such as fixed-time scheduling and empirical rule-based scheduling. These methods are simple to implement, but they have limitations in dealing with dynamic changes in water demand and complex operating conditions. In recent years, some researchers have begun to apply optimization algorithms to water supply scheduling, such as genetic algorithms, particle swarm optimization, and NSGA-II multi-objective optimization algorithms.

[0004] However, existing technologies still have room for improvement in the following aspects: Traditional optimization algorithms typically assume that water demand is deterministic or follows a known probability distribution, while in reality, water demand is affected by various factors such as weather, holidays, and emergencies, exhibiting randomness and uncertainty. Existing methods often employ a single time scale for scheduling, making it difficult to coordinate the relationship between long-term strategic planning and short-term real-time control. Long-term scheduling needs to consider factors such as reservoir storage and energy costs, while short-term scheduling requires rapid response to real-time demand changes; the two differ in time scale and decision-making objectives. Traditional methods have limited consideration for coordinating the interests of multiple stakeholders. Water supply systems involve multiple stakeholders, including managers, multiple water-using areas, and different water sources; the objective functions of these stakeholders may conflict, necessitating the establishment of an effective coordination mechanism. Research on the application of deep reinforcement learning in water supply scheduling is under development; existing research offers limited theoretical guarantees for algorithm convergence, and the multi-objective weights typically use fixed values, making it difficult to adapt to dynamically changing operating environments.

[0005] Therefore, there is an urgent need for an intelligent water supply scheduling and control method based on a multi-objective algorithm to solve the problems of insufficient handling of water demand uncertainty, poor coordination among multiple water sources, and weak short-term control stability in existing technologies. Summary of the Invention

[0006] To address these issues, this invention provides an intelligent water supply scheduling and control method and apparatus based on a multi-objective algorithm, which solves the problems of insufficient handling of water demand uncertainty, poor coordination among multiple water sources, weak short-term control stability, and difficulty in achieving robust multi-objective optimization scheduling in existing water supply scheduling methods.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent water supply scheduling and control method based on a multi-objective algorithm, characterized in that it includes:

[0008] The water supply network is equipped with sensors to collect real-time operational status data, including node pressure, pipeline flow rate, water tank level, pump power, and valve opening.

[0009] Based on the operational status data and combined with historical water usage data, water demand forecasts are performed at multiple time scales to generate water demand forecast sequences for long-term, medium-term, and short-term time scales.

[0010] Based on the long-term water demand prediction sequence, a multi-bar optimization model considering the uncertainty of water demand probability distribution is constructed and solved to obtain the long-term water storage and release operation strategy of the reservoir.

[0011] Based on the water demand forecast sequence at a medium-term scale and the long-term water storage and release operation strategy of the reservoir, a Stackelberg dynamic game model involving multiple stakeholders is constructed and the Stackelberg equilibrium is solved to obtain a multi-source joint scheduling scheme.

[0012] Based on the short-term water demand prediction sequence, the operating status data, and the multi-water source joint scheduling scheme, an improved multi-agent deep deterministic policy gradient algorithm that integrates Lyapunov stability constraints and adaptive Pareto front evolution mechanism is used to generate dynamic adjustment strategies for pump speed and valve opening.

[0013] Based on the dynamic adjustment strategy, corresponding optimized scheduling instructions are generated and sent to the execution equipment of the water supply network to complete the overall intelligent scheduling and control of the water supply system.

[0014] As a preferred scheme for intelligent water supply scheduling and control methods based on multi-objective algorithms, a decomposition-integration prediction approach is adopted in the process of multi-time-scale water demand prediction; the steps of multi-time-scale water demand prediction are as follows:

[0015] The water demand time series is decomposed into trend, periodic and random components;

[0016] The trend term is predicted using a long short-term memory neural network to obtain the trend term prediction result; the periodic term is modeled using Fourier series expansion to obtain the periodic term prediction result; and the random term is quantified for uncertainty using Gaussian process regression to obtain the random term prediction result.

[0017] The prediction results of the trend term, the periodic term, and the random term are integrated to obtain the predicted water demand and confidence interval for a future set time.

[0018] As a preferred scheme of intelligent water supply scheduling and control method based on multi-objective algorithm, in the process of constructing the sub-Bruker optimization model that considers the uncertainty of water demand probability distribution, a fuzzy set is generated by the historical mean vector of water demand and the upper bound of the covariance matrix; the uncertainty of water demand probability distribution is described by the fuzzy set; the objective of the sub-Bruker optimization model is to minimize the expected system operating cost under the worst distribution.

[0019] The expression for the sub-Brussels bar optimization model is:

[0020]

[0021] In the formula, This is a vector of decision variables, including the water release volume of each reservoir; This is the feasible decision domain; Water demand is a random variable The probability distribution; This is a function representing the system's operating cost. Represents the distribution Expectation calculation;

[0022] The expression for the fuzzy set is:

[0023]

[0024] In the formula, To support the collection The set of all probability measures on; This is a vector of historical average water demand. This is the upper bound of the covariance matrix of water demand; Represents a semi-positive definite order relation;

[0025] In solving the sub-Brubars optimization model, the sub-Brubars optimization problem is transformed into a deterministic equivalent form for solution using semidefinite programming duality theory.

[0026] As a preferred scheme of intelligent water supply scheduling and control method based on multi-objective algorithm, the leader decision variables of the Stackelberg dynamic game model are water price vector and water supply volume; the follower decision variables are water consumption; the Stackelberg equilibrium is solved by backward induction to obtain the optimal water price and water supply strategy.

[0027] The utility function expression for the follower is:

[0028]

[0029] In the formula, This represents the water user satisfaction coefficient. The water price is for the corresponding water-using area; This is the water usage deviation penalty coefficient; For reference water consumption; For followers Water consumption decision variables.

[0030] As a preferred embodiment of a smart water supply scheduling and control method based on a multi-objective algorithm, the construction process of the improved multi-agent deep deterministic policy gradient algorithm is as follows:

[0031] Each pump and valve in the water supply system is defined as an independent intelligent agent, and a multi-agent Markov decision process is constructed.

[0032] Based on the multi-agent Markov decision process, a system state space containing pipeline operation parameters and time characteristics is defined, as well as a continuous action space of the agents corresponding to the pump speed ratio and valve opening percentage.

[0033] Based on the system state space and the action space, a multi-objective reward function is designed that takes into account energy consumption, water pressure stability, water supply reliability and action smoothness.

[0034] Based on the aforementioned multi-objective reward function, the basic framework of the algorithm is built using an Actor-Critic architecture;

[0035] Based on the aforementioned framework, Lyapunov stability constraints are introduced to ensure stable convergence of the algorithm's policy update process.

[0036] Based on the algorithm framework that incorporates the Lyapunov stability constraint, an adaptive Pareto front evolution mechanism is introduced to dynamically balance the weights of multiple objectives, thereby obtaining the improved multi-agent deep deterministic policy gradient algorithm.

[0037] The expression for the multi-objective reward function is:

[0038]

[0039] In the formula, These are the weighting coefficients; Rewards for stress stability; As an energy consumption reward; Rewards for fulfilling demand; Reward for smooth motion; As a comprehensive reward;

[0040] The expression for the Lyapunov stability constraint is:

[0041]

[0042]

[0043] In the formula, Let t be the system state The corresponding Lyapunov function value; The system state at time t+1 The corresponding Lyapunov function value; The control action output by the agent at time t; This is the convergence rate parameter; This represents the desired equilibrium state. For nodes exist Constant pressure head; For nodes Reference pressure head; For water tank exist The water level at that moment; For water tank Reference water level; This represents the total number of water tanks in the pipeline network.

[0044] The update formula for the multi-objective weights is:

[0045]

[0046] In the formula, For temperature parameters; For the first During the nth iteration The weight of each objective; For the first The crowded distance to each target; The target total number.

[0047] This invention also provides an intelligent water supply scheduling and control device based on a multi-objective algorithm, employing the above-mentioned intelligent water supply scheduling and control method based on a multi-objective algorithm, comprising:

[0048] The data acquisition module is used to collect and acquire real-time operating status data of the water supply network through sensor acquisition devices deployed in the water supply network; the operating status data includes node pressure, pipeline flow rate, water tank level, water pump power, and valve opening.

[0049] The multi-timescale demand forecasting module is used to forecast water demand at multiple time scales based on the operational status data and historical water use data, generating water demand forecast sequences for long-term, medium-term, and short-term time scales.

[0050] The sub-Blu-bar optimization model construction and solution module is used to construct and solve a sub-Blu-bar optimization model that considers the uncertainty of the probability distribution of water demand based on the long-term water demand prediction sequence, so as to obtain the long-term water storage and release operation strategy of the reservoir.

[0051] The Stackelberg dynamic game model optimization module is used to construct a multi-agent Stackelberg dynamic game model and solve the Stackelberg equilibrium based on the medium-term water demand prediction sequence and the long-term water storage and release operation strategy of the reservoir, so as to obtain a multi-source joint scheduling scheme.

[0052] The deep reinforcement learning decision module is used to generate dynamic adjustment strategies for pump speed and valve opening based on the short-term water demand prediction sequence, the operating status data and the multi-water source joint scheduling scheme. It uses an improved multi-agent deep deterministic policy gradient algorithm that integrates Lyapunov stability constraints and adaptive Pareto front evolution mechanism.

[0053] The scheduling and execution module is used to generate corresponding optimized scheduling instructions based on the dynamic adjustment strategy and send them to the execution equipment of the water supply network to complete the overall intelligent scheduling and control of the water supply system.

[0054] As a preferred embodiment of an intelligent water supply scheduling and control device based on a multi-objective algorithm, the multi-time-scale demand forecasting module employs a decomposition-integration forecasting approach during the multi-time-scale water demand forecasting process. The steps for multi-time-scale water demand forecasting are as follows:

[0055] The water demand time series is decomposed into trend, periodic and random components;

[0056] The trend term is predicted using a long short-term memory neural network to obtain the trend term prediction result; the periodic term is modeled using Fourier series expansion to obtain the periodic term prediction result; and the random term is quantified for uncertainty using Gaussian process regression to obtain the random term prediction result.

[0057] The prediction results of the trend term, the periodic term, and the random term are integrated to obtain the predicted water demand and confidence interval for a future set time.

[0058] As a preferred solution for an intelligent water supply scheduling and control device based on a multi-objective algorithm, the sub-Bruker optimization model construction and solution module generates a fuzzy set by using the historical mean vector of water demand and the upper bound of the covariance matrix during the construction of the sub-Bruker optimization model that considers the uncertainty of the probability distribution of water demand. The fuzzy set is used to describe the uncertainty of the probability distribution of water demand. The objective of the sub-Bruker optimization model is to minimize the expected system operating cost under the worst-case distribution.

[0059] The expression for the sub-Brussels bar optimization model is:

[0060]

[0061] In the formula, This is a vector of decision variables, including the water release volume of each reservoir; This is the feasible decision domain; Water demand is a random variable The probability distribution; This is a function representing the system's operating cost. Represents the distribution Expectation calculation;

[0062] The expression for the fuzzy set is:

[0063]

[0064] In the formula, To support the collection The set of all probability measures on; This is a vector of historical average water demand. This is the upper bound of the covariance matrix of water demand; Represents a semi-positive definite order relation;

[0065] In solving the sub-Brubars optimization model, the sub-Brubars optimization problem is transformed into a deterministic equivalent form for solution using semidefinite programming duality theory.

[0066] As a preferred solution for an intelligent water supply scheduling and control device based on a multi-objective algorithm, in the Stackelberg dynamic game model optimization module, the leader's decision variables in the Stackelberg dynamic game model are the water price vector and the water supply volume; the follower's decision variables are the water consumption; the Stackelberg equilibrium is solved by backward induction to obtain the optimal water price and water supply strategy.

[0067] The utility function expression for the follower is:

[0068]

[0069] In the formula, This represents the water user satisfaction coefficient. The water price is for the corresponding water-using area; This is the water usage deviation penalty coefficient; For reference water consumption; For followers Water consumption decision variables.

[0070] As a preferred embodiment of an intelligent water supply scheduling and control device based on a multi-objective algorithm, the construction process of the improved multi-agent deep deterministic policy gradient algorithm in the deep reinforcement learning decision module is as follows:

[0071] Each pump and valve in the water supply system is defined as an independent intelligent agent, and a multi-agent Markov decision process is constructed.

[0072] Based on the multi-agent Markov decision process, a system state space containing pipeline operation parameters and time characteristics is defined, as well as a continuous action space of the agents corresponding to the pump speed ratio and valve opening percentage.

[0073] Based on the system state space and the action space, a multi-objective reward function is designed that takes into account energy consumption, water pressure stability, water supply reliability and action smoothness.

[0074] Based on the aforementioned multi-objective reward function, the basic framework of the algorithm is built using an Actor-Critic architecture;

[0075] Based on the aforementioned framework, Lyapunov stability constraints are introduced to ensure stable convergence of the algorithm's policy update process.

[0076] Based on the algorithm framework that incorporates the Lyapunov stability constraint, an adaptive Pareto front evolution mechanism is introduced to dynamically balance the weights of multiple objectives, thereby obtaining the improved multi-agent deep deterministic policy gradient algorithm.

[0077] The expression for the multi-objective reward function is:

[0078]

[0079] In the formula, These are the weighting coefficients; Rewards for stress stability; As an energy consumption reward; Rewards for fulfilling demand; Reward for smooth motion; As a comprehensive reward;

[0080] The expression for the Lyapunov stability constraint is:

[0081]

[0082]

[0083] In the formula, Let t be the system state The corresponding Lyapunov function value; The system state at time t+1 The corresponding Lyapunov function value; The control action output by the agent at time t; This is the convergence rate parameter; This represents the desired equilibrium state. For nodes exist Constant pressure head; For nodes Reference pressure head; For water tank exist The water level at that moment; For water tank Reference water level; This represents the total number of water tanks in the pipeline network.

[0084] The update formula for the multi-objective weights is:

[0085]

[0086] In the formula, For temperature parameters; For the first During the nth iteration The weight of each objective; For the first The crowded distance to each target; The target total number.

[0087] The present invention has the following advantages:

[0088] First, this invention constructs a hierarchical scheduling framework with three time scales: long-term, medium-term, and short-term, coordinating decision-making problems at different time scales. At the long-term scale, a distributed bar optimization approach is used to handle reservoir water storage and release strategies; at the medium-term scale, Stackelberg game theory is used to coordinate joint scheduling of multiple water sources; and at the short-term scale, deep reinforcement learning is used to achieve real-time control, forming a scheduling decision chain.

[0089] Secondly, this invention introduces the sub-Bruker optimization theory to address the uncertainty of water demand. By constructing fuzzy sets to describe the uncertainty of the probability distribution, the optimal decision under the worst distribution is obtained, which helps to improve the robustness of the scheduling scheme.

[0090] Third, this invention establishes a Stackelberg dynamic game model to coordinate the interests of multiple stakeholders. Through a leader-follower hierarchical decision-making structure, it realizes the interaction between water pricing and water consumption response, which helps to promote the overall efficiency of the water supply system.

[0091] Fourth, this invention proposes an improved multi-agent deep deterministic policy gradient algorithm, which introduces Lyapunov stability theory to ensure the convergence of the algorithm. By designing Lyapunov functions and constraints, it helps to ensure that the system state remains stable during policy updates.

[0092] Fifth, this invention designs an adaptive Pareto front evolution mechanism, which dynamically adjusts the multi-objective weights according to the distribution density of the Pareto front, guiding the strategy search to explore sparse regions, which helps to improve the quality and diversity of solution sets for multi-objective optimization.

[0093] Sixth, this invention uses polynomial chaotic expansion for uncertainty quantification and propagation analysis, which can calculate the output statistical moments and provide a basis for robust scheduling decisions. Attached Figure Description

[0094] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0095] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0096] Figure 1 This is a flowchart illustrating an intelligent water supply scheduling and control method based on a multi-objective algorithm provided in Embodiment 1 of the present invention.

[0097] Figure 2 This is a schematic diagram of Lyapunov stability constraints in an intelligent water supply scheduling and control method based on a multi-objective algorithm provided in Embodiment 1 of the present invention.

[0098] Figure 3 This is a schematic diagram of the adaptive Pareto front evolution mechanism in an intelligent water supply scheduling and control method based on a multi-objective algorithm provided in Embodiment 1 of the present invention;

[0099] Figure 4 This is a schematic diagram of the architecture of an intelligent water supply scheduling and control device based on a multi-objective algorithm provided in Embodiment 2 of the present invention. Detailed Implementation

[0100] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0101] Example 1

[0102] See Figure 1 Embodiment 1 of the present invention provides an intelligent water supply scheduling and control method based on a multi-objective algorithm, comprising the following steps:

[0103] S1. Real-time acquisition of operational status data of the water supply network through sensor acquisition devices deployed in the water supply network; the operational status data includes node pressure, pipeline flow rate, water tank level, water pump power, and valve opening.

[0104] S2. Based on the operational status data and combined with historical water usage data, perform multi-timescale water demand forecasting to generate water demand forecast sequences for long-term, medium-term, and short-term timescales.

[0105] S3. Based on the long-term water demand prediction sequence, construct a sub-Bruker optimization model that considers the uncertainty of the probability distribution of water demand and solve it to obtain the long-term water storage and release operation strategy of the reservoir.

[0106] S4. Based on the water demand prediction sequence at the medium-term scale and the long-term water storage and release operation strategy of the reservoir, construct a Stackelberg dynamic game model involving multiple stakeholders and solve the Stackelberg equilibrium to obtain a multi-source joint scheduling scheme.

[0107] S5. Based on the short-term water demand prediction sequence, the operating status data, and the multi-water source joint scheduling scheme, an improved multi-agent deep deterministic policy gradient algorithm that integrates Lyapunov stability constraints and adaptive Pareto front evolution mechanism is used to generate a dynamic adjustment strategy for pump speed and valve opening.

[0108] S6. Based on the dynamic adjustment strategy, generate corresponding optimized scheduling instructions and send them to the execution equipment of the water supply network to complete the overall intelligent scheduling and control of the water supply system.

[0109] In this embodiment, in step S1, the operating status data of the water supply network is collected in real time through the sensing and acquisition equipment deployed in the water supply network; the operating status data includes node pressure, pipeline flow rate, water tank level, water pump power and valve opening.

[0110] Specifically, the system uses SCADA systems, smart water meters, pressure sensors, flow meters, and other equipment to collect real-time operational status data of the water supply network, with a collection frequency of once per minute. The collected data includes pressure at each node, flow rate in each pipe, water tank level, pump power, valve opening, etc.

[0111] In this embodiment, in step S2, based on the operating status data and combined with historical water use data, water demand prediction is performed at multiple time scales to generate water demand prediction sequences for three time scales: long-term, medium-term, and short-term.

[0112] Specifically, multi-timescale demand forecasting employs a decomposition-integration forecasting framework. The steps for multi-timescale water demand forecasting are as follows:

[0113] Empirical mode decomposition is used to decompose the water demand time series into trend, periodic and random components.

[0114] The trend term is predicted using a Long Short-Term Memory (LSTM) neural network, consisting of two LSTM layers (128 neurons each) and two fully connected layers. The periodic term is modeled using Fourier series expansion.

[0115]

[0116] In the formula, For constant terms; and These are the Fourier coefficients; The period length; It represents the harmonic order.

[0117] Gaussian process regression is used to quantify the uncertainty of the random term, and the squared exponential kernel function is selected:

[0118]

[0119] In the formula, The signal variance; This is the length scale parameter.

[0120] Integrating the three predictions yields the future. The forecast values ​​for water demand at each time step and their 95% confidence intervals are provided. The forecast duration for the long-term scale is 168 to 720 hours, the forecast duration for the medium-term scale is 24 to 168 hours, and the forecast duration for the short-term scale is 1 to 24 hours.

[0121] In this embodiment, in step S3, based on the long-term water demand prediction sequence, a sub-Bruker optimization model considering the uncertainty of the probability distribution of water demand is constructed and solved to obtain the long-term water storage and release operation strategy of the reservoir.

[0122] Specifically, on a long-term scale, a reservoir water storage and release optimization model is established based on the bibliometric optimization theory. A fuzzy set is defined as the set of all possible water demand probability distributions, generated by moment information constraints from historical data.

[0123]

[0124] In the formula, To support the collection The set of all probability measures on; This is a vector of historical average water demand. This is the upper bound of the covariance matrix of water demand; This indicates a semi-positive definite order relation;

[0125] Establish a multi-bar optimization model:

[0126]

[0127] In the formula, This is a vector of decision variables, including the water release volume of each reservoir; This is the feasible decision domain; Water demand is a random variable The probability distribution; This is a function representing the system's operating cost. Represents the distribution The expected operation.

[0128] The above sub-Bruker optimization problem can be transformed into a deterministic equivalent form using semidefinite programming duality theory:

[0129]

[0130] In the formula, For symmetric matrix variables; For vector variables; It is a scalar variable; Let be the trace of the matrix.

[0131] In this embodiment, in step S4, based on the medium-term water demand prediction sequence and the long-term water storage and release operation strategy of the reservoir, a multi-agent Stackelberg dynamic game model is constructed and the Stackelberg equilibrium is solved to obtain a multi-source joint scheduling scheme.

[0132] Specifically, a Stackelberg dynamic game model with multiple participants is constructed, where the set of game participants is defined. Participant 0 is the water supply system manager, i.e., the leader. to for Each water-using area is a follower.

[0133] The leader's decision variables are the water price vector. and the water supply from each water source Followers The utility function is defined as:

[0134]

[0135] In the formula, This represents the water user satisfaction coefficient. The water price is for the corresponding water-using area; This is the water usage deviation penalty coefficient; For reference water consumption; For followers Water consumption decision variables.

[0136] Followers The optimization problem is:

[0137]

[0138] Through the first-order optimality condition Find the optimal response function of the follower:

[0139]

[0140] The leader's objective function is to maximize the overall system benefit:

[0141]

[0142]

[0143] In the formula, It is a linear water supply cost function; For water source The unit water supply cost; This is the supply and demand balance penalty coefficient.

[0144] In this embodiment, backward induction is used to solve for the Stackelberg equilibrium:

[0145] First, for a given water price Calculate the optimal response of each follower. Then, the optimal response function is substituted into the leader's objective function to solve for the leader's optimal strategy. Obtain the Stackelberg equilibrium. .

[0146] In this embodiment, in step S5, based on the short-term water demand prediction sequence, the operating status data, and the multi-water source joint scheduling scheme, an improved multi-agent deep deterministic policy gradient algorithm that integrates Lyapunov stability constraints and adaptive Pareto front evolution mechanism is used to generate a dynamic adjustment strategy for pump speed and valve opening.

[0147] Specifically, real-time control is achieved using an improved multi-agent deep deterministic policy gradient algorithm. The construction process of the improved multi-agent deep deterministic policy gradient algorithm is as follows:

[0148] T1. Define each pump and valve in the water supply system as an agent and construct a multi-agent Markov decision process.

[0149] T2. Based on the multi-agent Markov decision process, define the system state space containing pipeline operation parameters and time characteristics, and the agent continuous action space corresponding to the pump speed ratio and valve opening percentage.

[0150] The system state space is defined as follows:

[0151]

[0152] In the formula, for Pressure head vectors at each node at any given time; for The flow vector of each pipe at any given time; for Water demand vector at each node at any given time; for Control action vector at any given moment; This is the current hour. This is the current week number.

[0153] The action space of the intelligent agent is a continuous control variable. For a water pump, the action represents the speed ratio, with a value range of [missing value]. For valves, the action represents the percentage of opening, with a value range of [value range missing]. .

[0154] T3. Based on the system state space and the action space, design a multi-objective reward function that takes into account energy consumption, water pressure stability, water supply reliability and action smoothness.

[0155] The expression for the multi-objective reward function is as follows:

[0156]

[0157] In the formula, The weighting coefficients are set to their initial values. ; Rewards for stress stability; As an energy consumption reward; Rewards for fulfilling demand; Reward for smooth motion; For comprehensive rewards.

[0158] The expression for the pressure stability reward is:

[0159]

[0160] In the formula, The total number of nodes; For nodes exist Constant pressure head; and They are nodes The minimum and maximum permissible pressure head, in this embodiment, , This refers to the allowable range of pressure head.

[0161] The expression for the energy consumption reward is:

[0162]

[0163] In the formula, The density of water; It is the acceleration due to gravity; For water pumps The efficiency ranges from 0.75 to 0.85; This represents the total number of water pumps in the system. Let be the water flow rate of the k-th pump at time t; Let be the head of the k-th pump at time t.

[0164] The expression for the reward that satisfies the demand is:

[0165]

[0166] In the formula, This represents the total number of pipeline nodes. Let be the water demand of node j at time t; Let be the actual water supply of node j at time t.

[0167] The expression for the smooth reward of the action is:

[0168]

[0169] In the formula, The total number of agents; This refers to the control action of the i-th agent at time t. Let be the control action of the i-th agent at time t-1.

[0170] T4. Based on the multi-objective reward function, the basic framework of the algorithm is built using the Actor-Critic architecture.

[0171] T5. Based on the aforementioned basic framework, Lyapunov stability constraints are introduced to ensure stable convergence of the algorithm's policy update process.

[0172] Specifically, such as Figure 2 As shown, define the Lyapunov function:

[0173]

[0174] In the formula, Let t be the system state The corresponding Lyapunov function value; For nodes exist Constant pressure head; For nodes Reference pressure head; For water tank exist The water level at that moment; For water tank Reference water level.

[0175] Design Lyapunov constraints:

[0176]

[0177] In the formula, The system state at time t+1 The corresponding Lyapunov function value; The control action output by the agent at time t; This is the convergence rate parameter; This represents the desired equilibrium state.

[0178] When updating the policy, solve the following constrained optimization problem:

[0179]

[0180] The constrained optimization problem is transformed into an unconstrained problem using the Lagrange multiplier method.

[0181]

[0182] The policy parameters are updated jointly using the primal-dual gradient descent method. and Lagrange multipliers :

[0183]

[0184]

[0185] In the formula, and The learning rates are the policy and the multiplier, respectively.

[0186] T6. Based on the algorithm framework that incorporates the Lyapunov stability constraint, an adaptive Pareto front evolution mechanism is introduced to dynamically balance the weights of multiple objectives, thereby obtaining the improved multi-agent deep deterministic policy gradient algorithm.

[0187] Specifically, such as Figure 3 As shown, the embodiment designs an adaptive Pareto front evolution mechanism. The Pareto dominance relation for the multi-objective optimization problem is defined as follows: for two solutions... and If the following conditions are met:

[0188]

[0189] Then it is called Pareto dominance .

[0190] Define the Pareto front as the set of all non-dominated solutions:

[0191]

[0192] Design an adaptive weight adjustment strategy. Calculate the... Crowded distance between targets:

[0193]

[0194] In the formula, For the first The average of the targets.

[0195] Update weights based on crowding distance:

[0196]

[0197] In the formula, For temperature parameters, ; For the first During the nth iteration The weight of each objective; For the first The crowded distance to each target; The target total number.

[0198] By adaptively adjusting the weights, the policy search is guided to explore the sparse region of the Pareto front, thereby increasing the diversity of the solution set. The weights are updated every 100 iterations.

[0199] In this embodiment, in step S6, based on the dynamic adjustment strategy, a corresponding optimized scheduling instruction is generated and sent to the execution equipment of the water supply network to complete the overall intelligent scheduling and control of the water supply system.

[0200] Specifically, the real-time generated dynamic adjustment strategies for pump speed, valve opening, etc., are converted into optimized scheduling instructions that conform to the protocol specifications of the water supply network control system, including pump frequency control instructions and valve opening adjustment instructions. These instructions are then sent to the corresponding pumps, valves, and other actuators via the water supply network's communication network. Upon receiving the instructions, each actuator adjusts its operating status according to the set parameters, achieving dynamic control of the water supply network's pressure and flow. Simultaneously, continuous data collection of equipment operation feedback forms a closed-loop control system, ensuring the overall stability and reliability of the water supply system and completing the intelligent scheduling and control process.

[0201] In this embodiment, polynomial chaotic expansion is also used for uncertainty quantification and propagation analysis.

[0202] Modeling water demand uncertainty as a random variable Hermite polynomials are used as the orthogonal basis:

[0203]

[0204] The polynomial chaotic expansion output by the model is:

[0205]

[0206] Number of expanded terms Based on the dimensions of the input random variables and polynomial order Confirmed. This embodiment takes... .

[0207] The expansion coefficients are calculated using the sparse grid locus of points method. For a one-dimensional standard normal distribution, the locus of points and weights are determined using the Gauss-Hermite quadrature formula. For the multidimensional case, the locus of points is constructed using a Smolyak sparse grid.

[0208] The statistical moments of the output are calculated using polynomial chaotic expansion:

[0209]

[0210] In the formula, the inner product of the Hermite polynomials

[0211] Uncertainty is propagated to scheduling decisions to generate robust scheduling schemes. For each candidate scheduling scheme, the mean and variance of its objective function value are calculated, and the scheme with the smaller mean and smaller variance is selected.

[0212] In this embodiment, a real-time verification mechanism for the hydraulic model is also established. The hydraulic equations of the water supply network include nodal continuity equations and pipeline energy equations.

[0213] The nodal continuity equation is as follows:

[0214]

[0215] The pipeline energy equation is:

[0216]

[0217] In the formula, the pipe resistance coefficient Calculated using the Hazen-Williams formula:

[0218]

[0219] In the formula, This refers to the length of the pipe. The roughness coefficient is the Hazen-Williams roughness coefficient. This refers to the pipe diameter.

[0220] The pump head-flow rate relationship was fitted using a quadratic curve:

[0221]

[0222] In the formula, This represents the coefficient of the pump characteristic curve.

[0223] Valve head loss is calculated using the following formula:

[0224]

[0225] In the formula, This is the local loss coefficient of the valve; This represents the cross-sectional area of ​​the pipe.

[0226] The Newton-Raphson method is used to solve the above nonlinear equation system. The residual vector is defined. The iterative update formula is:

[0227]

[0228] In the formula, It is a Jacobian matrix.

[0229] Real-time monitoring data is compared with model calculation results, and automatic model parameter verification is triggered when the pressure error exceeds 5% or the flow error exceeds 10%. The least squares method is used to calibrate the pipe roughness coefficient and the node water demand correction coefficient.

[0230] The scheduling and execution module will optimize the dispatching instructions and send them to the execution equipment. For water pumps, the speed setpoint is sent to the frequency converter; for valves, the opening setpoint is sent to the electric actuator. The instruction is sent every 5 minutes.

[0231] In one possible implementation, the application was validated using a city's water supply network. This network comprises 3 water treatment plants, 5 booster pump stations, 12 water tanks, 156 nodes, and 203 pipelines. Data was collected every minute, with a historical data period of 3 years.

[0232] Experimental results show that, compared with the traditional NSGA-II algorithm, the method in this embodiment reduces energy consumption by 18.3%, improves stress stability by 32.7%, and increases demand satisfaction rate by 8.5%. Compared with fixed-weight multi-objective reinforcement learning methods, the adaptive Pareto front evolution mechanism in this embodiment expands the coverage of the Pareto front by 45.2% and improves the diversity of solution sets.

[0233] In one possible embodiment, the network structure of the deep reinforcement learning algorithm is further optimized.

[0234] The policy network employs a three-layer fully connected structure, with 256 neurons in each layer, and uses ReLU activation. The output layer uses the tanh activation function to map actions to... The interval is then mapped to the actual action range through a linear transformation.

[0235] The Q-network employs a dual-criteria structure, with two identical but independent critic networks to mitigate the Q-value overestimation problem. Each critic network receives a concatenated vector of state and action as input, which outputs a scalar Q-value after passing through three fully connected layers.

[0236] The target network employs a soft update strategy:

[0237]

[0238]

[0239] In the formula, and These are the parameters for the target policy network and the target Q network; This is the soft update coefficient.

[0240] The experience replay buffer capacity is The batch size is 256. Noise exploration is performed using the Ornstein-Uhlenbeck process:

[0241]

[0242] In the formula, The mean regression rate; The mean; The diffusion coefficient is denoted as . This is the Wiener process.

[0243] During training, a cosine annealing strategy is used for the learning rate:

[0244]

[0245] In the formula, The initial learning rate; Minimum learning rate; This represents the total number of training steps.

[0246] In one possible embodiment, the present invention also adds anomaly detection and emergency dispatch functions.

[0247] Anomaly detection is performed using the Isolation Forest algorithm. For the state vector at each time step... Calculate its anomaly score:

[0248]

[0249]

[0250] In the formula, for Average path length across all isolated trees; Normalization factor; It is a harmonic number.

[0251] An anomaly alarm is triggered when the anomaly score exceeds the threshold of 0.7. The corresponding emergency dispatch plan is then activated based on the anomaly type.

[0252] In case of a pipe burst, quickly close the isolation valves in the affected area and adjust the surrounding water pumps to increase the water supply pressure;

[0253] If the water quality is abnormal, switch to a backup water source and increase the dosage of water treatment chemicals.

[0254] In the event of a power outage, backup generators will be activated to prioritize water supply to critical areas.

[0255] Emergency dispatch employs a combination of rule-based reasoning and reinforcement learning. A predefined emergency rule base is used, and a dedicated emergency dispatch strategy network is trained. In abnormal situations, emergency dispatch instructions are generated by combining the results of rule-based reasoning and the output of the strategy network.

[0256] The application scenarios of this invention are as follows:

[0257] In the daily scheduling scenarios of large-scale urban ring water supply networks, this invention can integrate the operation data of multiple water sources, multiple pumping stations, and multiple reservoirs, and achieve collaborative scheduling through multi-timescale prediction and game optimization. It can dynamically adjust the operating frequency of pumps and valve opening of each pumping station, balance the overall water pressure distribution of the network, and reduce the overall energy consumption of the system while ensuring the reliability of water supply at all times.

[0258] In the scenario of fluctuating water usage during morning and evening peak hours in residential communities, this invention can adjust the operating status of water supply pumping stations and secondary water supply equipment in advance based on real-time collected node flow and pressure data and short-term water demand forecast results. This can alleviate the problems of insufficient pressure at the end of the pipeline during peak hours and excessively high local pressure during off-peak hours, thereby improving the water usage experience for residents.

[0259] In the scenario of continuous production water supply in industrial parks, this invention can address the continuous water demand of industrial production, taking into account the stability of production water supply, pipeline pressure control and equipment operating energy consumption. Through multi-objective optimization, a scheduling scheme is generated to avoid the impact of water supply pressure fluctuations on the production process, while reducing industrial water supply costs.

[0260] In low-load water supply scenarios at night, this invention can combine long-term water storage and release strategies with short-term equipment optimization to adjust the reservoir's water storage capacity and the operation mode of the nighttime water supply pumping station, reduce unnecessary pump start-ups and shutdowns and inefficient operation, reduce nighttime energy consumption, reduce equipment mechanical wear, and extend equipment lifespan.

[0261] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0262] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0263] Example 2

[0264] See Figure 4 Embodiment 2 of the present invention also provides an intelligent water supply scheduling and control device based on a multi-objective algorithm, comprising:

[0265] The data acquisition module 001 is used to collect and acquire real-time operating status data of the water supply network through the sensing and acquisition equipment deployed in the water supply network; the operating status data includes node pressure, pipeline flow rate, water tank level, water pump power and valve opening.

[0266] The multi-timescale demand forecasting module 002 is used to perform multi-timescale water demand forecasting based on the operating status data and combined with historical water use data, and generate water demand forecasting sequences for three time scales: long-term, medium-term, and short-term.

[0267] The sub-Blu-bar optimization model construction and solution module 003 is used to construct and solve a sub-Blu-bar optimization model that considers the uncertainty of the probability distribution of water demand based on the long-term water demand prediction sequence, so as to obtain the long-term water storage and release operation strategy of the reservoir.

[0268] Stackelberg dynamic game model optimization module 004 is used to construct a multi-agent Stackelberg dynamic game model and solve the Stackelberg equilibrium based on the medium-term water demand prediction sequence and the long-term water storage and release operation strategy of the reservoir, so as to obtain a multi-source joint scheduling scheme.

[0269] The deep reinforcement learning decision module 005 is used to generate dynamic adjustment strategies for pump speed and valve opening based on the short-term scale water demand prediction sequence, the operating status data and the multi-water source joint scheduling scheme, by using an improved multi-agent deep deterministic policy gradient algorithm that integrates Lyapunov stability constraints and adaptive Pareto front evolution mechanism.

[0270] The scheduling execution module 006 is used to generate corresponding optimized scheduling instructions based on the dynamic adjustment strategy and send them to the execution equipment of the water supply network to complete the overall intelligent scheduling and control of the water supply system.

[0271] In this embodiment, the multi-timescale demand forecasting module 002 employs a decomposition-integration forecasting method during the multi-timescale water demand forecasting process; the steps of multi-timescale water demand forecasting are as follows:

[0272] The water demand time series is decomposed into trend, periodic and random components;

[0273] The trend term is predicted using a long short-term memory neural network to obtain the trend term prediction result; the periodic term is modeled using Fourier series expansion to obtain the periodic term prediction result; and the random term is quantified for uncertainty using Gaussian process regression to obtain the random term prediction result.

[0274] The prediction results of the trend term, the periodic term, and the random term are integrated to obtain the predicted water demand and confidence interval for a future set time.

[0275] In this embodiment, in the sub-Bruker optimization model construction and solution module 003, during the process of constructing the sub-Bruker optimization model that considers the uncertainty of the probability distribution of water demand, a fuzzy set is generated by the historical mean vector of water demand and the upper bound of the covariance matrix; the fuzzy set is used to describe the uncertainty of the probability distribution of water demand; the objective of the sub-Bruker optimization model is to minimize the expected system operating cost under the worst distribution.

[0276] The expression for the sub-Brussels bar optimization model is:

[0277]

[0278] In the formula, This is a vector of decision variables, including the water release volume of each reservoir; This is the feasible decision domain; Water demand is a random variable The probability distribution; This is a function representing the system's operating cost. Represents the distribution Expectation calculation;

[0279] The expression for the fuzzy set is:

[0280]

[0281] In the formula, To support the collection The set of all probability measures on; This is a vector of historical average water demand. This is the upper bound of the covariance matrix of water demand; This indicates a semi-positive definite order relation;

[0282] In solving the sub-Brubars optimization model, the sub-Brubars optimization problem is transformed into a deterministic equivalent form for solution using semidefinite programming duality theory.

[0283] In this embodiment, in the Stackelberg dynamic game model optimization module 004, the leader's decision variables in the Stackelberg dynamic game model are the water price vector and the water supply volume; the follower's decision variables are the water consumption; the Stackelberg equilibrium is solved by backward induction to obtain the optimal water price and water supply strategy.

[0284] The utility function expression for the follower is:

[0285]

[0286] In the formula, This represents the water user satisfaction coefficient. The water price is for the corresponding water-using area; This is the water usage deviation penalty coefficient; For reference water consumption; For followers Water consumption decision variables.

[0287] In this embodiment, the construction process of the improved multi-agent deep deterministic policy gradient algorithm in the deep reinforcement learning decision module 005 is as follows:

[0288] Each pump and valve in the water supply system is defined as an independent intelligent agent, and a multi-agent Markov decision process is constructed.

[0289] Based on the multi-agent Markov decision process, a system state space containing pipeline operation parameters and time characteristics is defined, as well as a continuous action space of the agents corresponding to the pump speed ratio and valve opening percentage.

[0290] Based on the system state space and the action space, a multi-objective reward function is designed that takes into account energy consumption, water pressure stability, water supply reliability and action smoothness.

[0291] Based on the aforementioned multi-objective reward function, the basic framework of the algorithm is built using an Actor-Critic architecture;

[0292] Based on the aforementioned framework, Lyapunov stability constraints are introduced to ensure stable convergence of the algorithm's policy update process.

[0293] Based on the algorithm framework that incorporates the Lyapunov stability constraint, an adaptive Pareto front evolution mechanism is introduced to dynamically balance the weights of multiple objectives, thereby obtaining the improved multi-agent deep deterministic policy gradient algorithm.

[0294] The expression for the multi-objective reward function is:

[0295]

[0296] In the formula, These are the weighting coefficients; Rewards for stress stability; As an energy consumption reward; Rewards for fulfilling demand; Reward for smooth motion; As a comprehensive reward;

[0297] The expression for the Lyapunov stability constraint is:

[0298]

[0299]

[0300] In the formula, Let t be the system state The corresponding Lyapunov function value; The system state at time t+1 The corresponding Lyapunov function value; The control action output by the agent at time t; This is the convergence rate parameter; This represents the desired equilibrium state. For nodes exist Constant pressure head; For nodes Reference pressure head; For water tank exist The water level at that moment; For water tank Reference water level; This represents the total number of water tanks in the pipeline network.

[0301] The update formula for the multi-objective weights is:

[0302]

[0303] In the formula, For temperature parameters; For the first During the nth iteration The weight of each objective; For the first The crowded distance to each target; The target total number.

[0304] It should be noted that the information interaction and execution process between the modules of the above system are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0305] Example 3

[0306] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code for a method of classifying and attributing abnormal temperature acquisition values ​​of an energy storage BMS. The program code includes instructions for executing the method of classifying and attributing abnormal temperature acquisition values ​​of an energy storage BMS according to Embodiment 1 or any possible implementation thereof.

[0307] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0308] Example 4

[0309] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0310] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can call the program instructions to execute a method for classifying and attributing abnormal temperature acquisition values ​​of an energy storage BMS according to Embodiment 1 or any possible implementation thereof.

[0311] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0312] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0313] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using program code executable by a computing system, thereby storing them in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0314] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A smart water supply scheduling and control method based on a multi-objective algorithm, characterized in that, include: The system uses sensor data acquisition devices deployed in the water supply network to collect real-time data on the network's operational status. The operational status data includes node pressure, pipeline flow rate, water tank level, water pump power, and valve opening. Based on the operational status data and combined with historical water usage data, water demand forecasts are performed at multiple time scales to generate water demand forecast sequences for long-term, medium-term, and short-term time scales. Based on the long-term water demand prediction sequence, a multi-bar optimization model considering the uncertainty of water demand probability distribution is constructed and solved to obtain the long-term water storage and release operation strategy of the reservoir. Based on the water demand forecast sequence at a medium-term scale and the long-term water storage and release operation strategy of the reservoir, a Stackelberg dynamic game model involving multiple stakeholders is constructed and the Stackelberg equilibrium is solved to obtain a multi-source joint scheduling scheme. Based on the short-term water demand prediction sequence, the operating status data, and the multi-water source joint scheduling scheme, an improved multi-agent deep deterministic policy gradient algorithm that integrates Lyapunov stability constraints and adaptive Pareto front evolution mechanism is used to generate dynamic adjustment strategies for pump speed and valve opening. Based on the dynamic adjustment strategy, corresponding optimized scheduling instructions are generated and sent to the execution equipment of the water supply network to complete the overall intelligent scheduling and control of the water supply system.

2. The intelligent water supply scheduling and control method based on a multi-objective algorithm according to claim 1, characterized in that, In the process of multi-timescale water demand forecasting, a decomposition-integration forecasting approach is adopted; the steps of multi-timescale water demand forecasting are as follows: The water demand time series is decomposed into trend, periodic and random components; The trend term is predicted using a long short-term memory neural network to obtain the trend term prediction result; The periodic terms are modeled using Fourier series expansion to obtain the prediction results for the periodic terms; The uncertainty of the random term is quantified by Gaussian process regression to obtain the prediction result of the random term; The prediction results of the trend term, the periodic term, and the random term are integrated to obtain the predicted water demand and confidence interval for a future set time.

3. The intelligent water supply scheduling and control method based on a multi-objective algorithm according to claim 2, characterized in that, In constructing the sub-Bruker optimization model that considers the uncertainty of the probability distribution of water demand, a fuzzy set is generated by the historical mean vector of water demand and the upper bound of the covariance matrix; the fuzzy set is used to describe the uncertainty of the probability distribution of water demand; the objective of the sub-Bruker optimization model is to minimize the expected system operating cost under the worst distribution. The expression for the sub-Brussels bar optimization model is: In the formula, This is a vector of decision variables, including the water release volume of each reservoir; This is the feasible decision domain; Water demand is a random variable The probability distribution; This is a function representing the system's operating cost. Represents the distribution Expectation calculation; The expression for the fuzzy set is: In the formula, To support the collection The set of all probability measures on; This is a vector of historical average water demand. This is the upper bound of the covariance matrix of water demand; Represents a semi-positive definite order relation; In solving the sub-Brubars optimization model, the sub-Brubars optimization problem is transformed into a deterministic equivalent form for solution using semidefinite programming duality theory.

4. The intelligent water supply scheduling and control method based on a multi-objective algorithm according to claim 3, characterized in that, The Stackelberg dynamic game model uses water price vector and water supply as the leader's decision variables, and water consumption as the follower's decision variables. The Stackelberg equilibrium is solved by backward induction to obtain the optimal water price and water supply strategy. The utility function expression for the follower is: In the formula, This represents the water user satisfaction coefficient. The water price is for the corresponding water-using area; This is the water usage deviation penalty coefficient; For reference water consumption; For followers Water consumption decision variables.

5. The intelligent water supply scheduling and control method based on a multi-objective algorithm according to claim 4, characterized in that, The construction process of the improved multi-agent deep deterministic policy gradient algorithm is as follows: Each pump and valve in the water supply system is defined as an independent intelligent agent, and a multi-agent Markov decision process is constructed. Based on the multi-agent Markov decision process, a system state space containing pipeline operation parameters and time characteristics is defined, as well as a continuous action space of the agents corresponding to the pump speed ratio and valve opening percentage. Based on the system state space and the action space, a multi-objective reward function is designed that takes into account energy consumption, water pressure stability, water supply reliability and action smoothness. Based on the aforementioned multi-objective reward function, the basic framework of the algorithm is built using an Actor-Critic architecture; Based on the aforementioned framework, Lyapunov stability constraints are introduced to ensure stable convergence of the algorithm's policy update process. Based on the algorithm framework that incorporates the Lyapunov stability constraint, an adaptive Pareto front evolution mechanism is introduced to dynamically balance the weights of multiple objectives, thereby obtaining the improved multi-agent deep deterministic policy gradient algorithm. The expression for the multi-objective reward function is: In the formula, These are the weighting coefficients; Rewards for stress stability; As an energy consumption reward; Rewards for fulfilling demand; Reward for smooth motion; As a comprehensive reward; The expression for the Lyapunov stability constraint is: In the formula, Let t be the system state The corresponding Lyapunov function value; The system state at time t+1 The corresponding Lyapunov function value; The control action output by the agent at time t; This is the convergence rate parameter; This represents the desired equilibrium state. For nodes exist Constant pressure head; For nodes Reference pressure head; For water tank exist The water level at that moment; For water tank Reference water level; This represents the total number of water tanks in the pipeline network. The update formula for the multi-objective weights is: In the formula, For temperature parameters; For the first During the nth iteration The weight of each objective; For the first The crowded distance to each target; The target total number.

6. An intelligent water supply scheduling and control device based on a multi-objective algorithm, employing the intelligent water supply scheduling and control method based on a multi-objective algorithm as described in any one of claims 1-5, characterized in that, include: The data acquisition module is used to collect and acquire real-time operational status data of the water supply network through sensor acquisition devices deployed in the water supply network. The operational status data includes node pressure, pipeline flow rate, water tank level, water pump power, and valve opening. The multi-timescale demand forecasting module is used to forecast water demand at multiple time scales based on the operational status data and historical water use data, generating water demand forecast sequences for long-term, medium-term, and short-term time scales. The sub-Blu-bar optimization model construction and solution module is used to construct and solve a sub-Blu-bar optimization model that considers the uncertainty of the probability distribution of water demand based on the long-term water demand prediction sequence, so as to obtain the long-term water storage and release operation strategy of the reservoir. The Stackelberg dynamic game model optimization module is used to construct a multi-agent Stackelberg dynamic game model and solve the Stackelberg equilibrium based on the medium-term water demand prediction sequence and the long-term water storage and release operation strategy of the reservoir, so as to obtain a multi-source joint scheduling scheme. The deep reinforcement learning decision module is used to generate dynamic adjustment strategies for pump speed and valve opening based on the short-term water demand prediction sequence, the operating status data and the multi-water source joint scheduling scheme. It uses an improved multi-agent deep deterministic policy gradient algorithm that integrates Lyapunov stability constraints and adaptive Pareto front evolution mechanism. The scheduling and execution module is used to generate corresponding optimized scheduling instructions based on the dynamic adjustment strategy and send them to the execution equipment of the water supply network to complete the overall intelligent scheduling and control of the water supply system.

7. The intelligent water supply scheduling and control device based on a multi-objective algorithm according to claim 6, characterized in that, In the multi-timescale demand forecasting module, a decomposition-integration forecasting method is adopted during the multi-timescale water demand forecasting process; the steps of multi-timescale water demand forecasting are as follows: The water demand time series is decomposed into trend, periodic and random components; The trend term is predicted using a long short-term memory neural network to obtain the trend term prediction result; The periodic terms are modeled using Fourier series expansion to obtain the prediction results for the periodic terms; The uncertainty of the random term is quantified by Gaussian process regression to obtain the prediction result of the random term; The prediction results of the trend term, the periodic term, and the random term are integrated to obtain the predicted water demand and confidence interval for a future set time.

8. The intelligent water supply scheduling and control device based on a multi-objective algorithm according to claim 7, characterized in that, In the sub-Bruker optimization model construction and solution module, during the process of constructing the sub-Bruker optimization model that considers the uncertainty of the probability distribution of water demand, a fuzzy set is generated by the historical mean vector of water demand and the upper bound of the covariance matrix; the fuzzy set is used to describe the uncertainty of the probability distribution of water demand; the objective of the sub-Bruker optimization model is to minimize the expected system operating cost under the worst distribution. The expression for the sub-Brussels bar optimization model is: In the formula, This is a vector of decision variables, including the water release volume of each reservoir; This is the feasible decision domain; Water demand is a random variable The probability distribution; This is a function representing the system's operating cost. Represents the distribution Expectation calculation; The expression for the fuzzy set is: In the formula, To support the collection The set of all probability measures on; This is a vector of historical average water demand. This is the upper bound of the covariance matrix of water demand; This indicates a semi-positive definite order relation; In solving the sub-Brubars optimization model, the sub-Brubars optimization problem is transformed into a deterministic equivalent form for solution using semidefinite programming duality theory.

9. The intelligent water supply scheduling and control device based on a multi-objective algorithm according to claim 8, characterized in that, In the Stackelberg dynamic game model optimization module, the leader's decision variables in the Stackelberg dynamic game model are the water price vector and the water supply volume; the follower's decision variables are the water consumption; the Stackelberg equilibrium is solved by backward induction to obtain the optimal water price and water supply strategy. The utility function expression for the follower is: In the formula, This represents the water user satisfaction coefficient. The water price is for the corresponding water-using area; This is the water usage deviation penalty coefficient; For reference water consumption; For followers Water consumption decision variables.

10. The intelligent water supply scheduling and control device based on a multi-objective algorithm according to claim 9, characterized in that, In the deep reinforcement learning decision module, the construction process of the improved multi-agent deep deterministic policy gradient algorithm is as follows: Each pump and valve in the water supply system is defined as an independent intelligent agent, and a multi-agent Markov decision process is constructed. Based on the multi-agent Markov decision process, a system state space containing pipeline operation parameters and time characteristics is defined, as well as a continuous action space of the agents corresponding to the pump speed ratio and valve opening percentage. Based on the system state space and the action space, a multi-objective reward function is designed that takes into account energy consumption, water pressure stability, water supply reliability and action smoothness. Based on the aforementioned multi-objective reward function, the basic framework of the algorithm is built using an Actor-Critic architecture; Based on the aforementioned framework, Lyapunov stability constraints are introduced to ensure stable convergence of the algorithm's policy update process. Based on the algorithm framework that incorporates the Lyapunov stability constraint, an adaptive Pareto front evolution mechanism is introduced to dynamically balance the weights of multiple objectives, thereby obtaining the improved multi-agent deep deterministic policy gradient algorithm. The expression for the multi-objective reward function is: In the formula, These are the weighting coefficients; Rewards for stress stability; As an energy consumption reward; Rewards for fulfilling demand; Reward for smooth motion; As a comprehensive reward; The expression for the Lyapunov stability constraint is: In the formula, Let t be the system state The corresponding Lyapunov function value; The system state at time t+1 The corresponding Lyapunov function value; The control action output by the agent at time t; This is the convergence rate parameter; This represents the desired equilibrium state. For nodes exist Constant pressure head; For nodes Reference pressure head; For water tank exist The water level at that moment; For water tank Reference water level; This represents the total number of water tanks in the pipeline network. The update formula for the multi-objective weights is: In the formula, For temperature parameters; For the first During the nth iteration The weight of each objective; For the first The crowded distance to each target; The target total number.

Images