New energy station operation decision support system based on multi-objective optimization

Abstract

The invention relates to the technical field of new energy power generation and control, and discloses a new energy station operation decision support system based on multi-objective optimization, which comprises a multi-source state space reconstruction module, a Riemannian manifold geometry engine module, a self-adaptive inertia Hamiltonian evolution module and a symplectic geometric integral and instruction mapping module. The system collects station data to construct a dimensionless state space, constructs a Riemannian metric tensor according to physical constraints to reconstruct a Riemannian manifold space, and calculates a geometric connection strength factor; the factor is used to adaptively modulate a virtual inertia matrix, and a dissipative Hamiltonian kinetic model is constructed; and finally, solving the steady-state generalized coordinates through a pungent-preserving numerical integration algorithm, and decoding the steady-state generalized coordinates into an equipment control instruction. According to the method, physical constraints are converted into geometric measurements, and a self-adaptive inertia mechanism is introduced, so that the problem of optimization convergence under multivariable strong constraints is solved, and the safety and accuracy of a control instruction are ensured.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation and control technology, specifically to a new energy power plant operation decision support system based on multi-objective optimization. Background Technology

[0002] With the increasing penetration rate of new energy sources, new energy power plants composed of wind power, photovoltaic, and energy storage units have become an important part of the power grid. In order to respond to grid dispatch instructions and ensure the economical operation of the power plants, it is necessary to coordinate the active and reactive power of each device within the power plant. This process is essentially a typical multivariable, strongly coupled, and multi-objective optimization problem constrained by complex physical constraints.

[0003] Existing operational decisions for new energy power plants typically rely on traditional mathematical programming methods or general heuristic algorithms. When dealing with hard physical constraints such as battery charge / discharge limits and inverter capacity limitations, these methods often employ external penalty functions or simple numerical projection methods. This approach performs reasonably well when the generalized coordinate vector is far from the constraint boundary, but as the system operating point approaches the critical value of the physical constraint, the gradient of the objective function often changes drastically. Due to a lack of awareness of the geometric topology of the solution space, conventional algorithms struggle to maintain numerical computational stability near the constraint boundary, easily leading to oscillations in control commands or direct limit violations, thus exposing equipment to safety risks.

[0004] Furthermore, existing optimization control strategies typically employ fixed iteration step sizes or inertia parameters. In multi-objective collaborative optimization, the curvature of the solution space varies across different regions. In nonlinear, high-curvature regions, fixed inertia parameters cannot adapt to rapid gradient changes, leading to slow algorithm convergence or failure to converge near the steady-state point. This algorithmic limitation makes it difficult for existing decision support systems to simultaneously achieve computational speed and control command accuracy in millisecond- or second-level real-time control. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a decision support system for the operation of new energy power stations based on multi-objective optimization. This system solves the problems of poor computational convergence and insufficient accuracy of control commands caused by the inability of algorithm parameters to be adaptively adjusted when dealing with complex physical hard constraints in existing new energy power station operation optimization methods.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a decision support system for the operation of new energy power plants based on multi-objective optimization.

[0007] The system includes a multi-source state space reconstruction module, a Riemannian manifold geometry engine module, an adaptive inertia Hamiltonian evolution module, and a symplectic geometric integral and instruction mapping module.

[0008] The multi-source state space reconstruction module is used to collect station operation data and scheduling instructions, and constructs an initial Euclidean state space using a dimensionless mapping method for physical quantities. This module maps physical variables of different dimensions within the station to a dimensionless generalized coordinate system, providing a data foundation for subsequent geometric transformations.

[0009] The Riemannian manifold geometry engine module connects to the multi-source state space reconstruction module to perform geometric reconstruction of the space. This module constructs a Riemannian metric tensor based on physical constraints, reconstructing the initial Euclidean state space into a Riemannian manifold space. During this process, the module calculates the geometric connection strength factor, which characterizes local topological features, using a geometric connection coefficient calculation method. This module transforms physical constraints into geometric curvature information of the state space, utilizing the spatial topology to constrain the evolution trajectory of generalized coordinate vectors.

[0010] The Adaptive Inertia Hamiltonian Evolution module, connected to the Riemannian Manifold Geometry Engine module, is used to construct the system's dynamic evolution model. This module calculates the virtual inertia matrix using a curvature-inertia mapping function and combines it with a multi-objective potential energy function to construct a dissipative Hamiltonian dynamic model including generalized momentum and generalized coordinates. This module establishes a mapping relationship between spatial curvature and the system's virtual mass, enabling adjustments to the system's dynamic response characteristics.

[0011] The symplectic geometric integral and command mapping module is connected to the adaptive inertia Hamiltonian evolution module for numerical solution and command generation. This module uses a symplectic structure numerical integration algorithm to iteratively solve the dissipative Hamiltonian dynamics model, obtains steady-state generalized coordinates, and decodes the steady-state generalized coordinates into control commands for various equipment in the new energy power station through an inverse mapping method.

[0012] Furthermore, in the aforementioned system, the components of the generalized coordinate vector constructed by the multi-source state-space reconstruction module include at least: a power variable characterizing the charging and discharging state of the energy storage unit, a curtailment variable characterizing the degree of active power reduction by the new energy power generation unit, and a reactive power variable characterizing the output of the power station's reactive power compensation device and inverter. The multi-objective potential energy function adopts a linear weighted sum form and includes at least: an economic cost potential energy term constructed based on the cycle life loss of the energy storage unit and the switching loss of the inverter, a network loss potential energy term constructed based on the power loss of the power collection lines and transformers within the power station, and a power tracking error potential energy term constructed based on the deviation between the actual total grid-connected power of the power station and the power dispatch reference command of the upper-level power grid.

[0013] The physical constraints handled by the Riemannian manifold geometry engine module include at least: the charging and discharging power range limitations of energy storage units, the maximum available power and non-negative curtailment of solar and wind power generation units due to meteorological resource constraints, and the circular boundary constraints of active and reactive power coupling of inverters due to rated capacity limitations. This module calculates the Riemannian metric tensor using a weighted summation formula of the identity matrix and the second derivative Hessian matrix of the logarithmic barrier function derived from the physical constraints. The Riemannian metric tensor exhibits the property that its eigenvalues ​​tend to infinity as the generalized coordinate vector approaches the constraint boundary, thus creating an increase in spatial metric distance at the constraint boundary. Furthermore, this module calculates the geometric connection strength factor using a formula for obtaining the Frobenius norm of the second kind using the inverse matrix and first-order partial derivatives of the Riemannian metric tensor, thereby quantifying the degree of geometric nonlinearity near the constraint boundary.

[0014] The adaptive inertia Hamiltonian evolution module calculates the virtual inertia matrix using a formula that modulates the gain of the fundamental inertia matrix with a hyperbolic tangent function that uses the geometric connection strength factor as the independent variable. This calculation ensures that the eigenvalues ​​of the virtual inertia matrix increase smoothly as the geometric connection strength factor increases, effectively increasing the virtual mass of the system near the constraint boundaries and suppressing oscillations in the state variables. The total energy function of the dissipative Hamiltonian dynamics model consists of the sum of the generalized kinetic energy term constructed from the inverse of the virtual inertia matrix and the generalized momentum, and the multi-objective potential energy function. The canonical equations of motion include the kinematic equations describing the rate of change of the generalized coordinates and the dynamic equilibrium equations describing the rate of change of the generalized momentum.

[0015] The symplectic geometric integral and command mapping module calculates the generalized momentum and generalized coordinates at the next time step using a symplectic Euler integral algorithm that includes explicit momentum update steps involving Hamiltonian function gradient forces and dissipation forces, as well as implicit coordinate update steps. This algorithm preserves the symplectic geometric structure of the system during iteration. Simultaneously, the module monitors the norm of the system's generalized velocity vector and determines whether the system has reached steady state by comparing the generalized velocity vector norm with a convergence threshold. When steady state is determined to have been reached, the module performs inverse mapping decoding, using reference parameters to reconstruct the steady-state generalized coordinates into active power control commands for the energy storage unit, active power reduction commands for the photovoltaic inverter, and reactive power control commands.

[0016] This invention provides a decision support system for the operation of new energy power plants based on multi-objective optimization. It has the following beneficial effects: 1. This invention transforms the physical inequality constraints of new energy power plants into Riemannian metric tensors in the state space using a Riemannian manifold geometry engine module. This method maps discontinuous hard constraint boundaries in Euclidean space to regions in Riemannian manifold space where metric values ​​tend to infinity, thus directly embedding the physical constraints into the geometric structure of the state space. This approach avoids the numerical truncation or feasible region overflow problems that occur at constraint boundaries in traditional optimization algorithms, ensuring that the generated energy storage charging / discharging and inverter power control commands are strictly within the physical safety operating range of the equipment.

[0017] 2. This invention utilizes an adaptive inertia Hamiltonian evolution module to establish a virtual inertia adjustment mechanism based on the geometric connection strength factor. The system can adjust the virtual inertia matrix in real time according to the spatial geometric curvature of its current position. It maintains a smaller inertia in flat regions far from constraints to improve convergence speed, and automatically increases the inertia in curved regions approaching constraints to enhance system damping. This mechanism effectively solves the problem of fixed-parameter algorithms struggling to balance computational speed and stability under strong nonlinear constraints, and suppresses numerical oscillations near the constraint boundaries.

[0018] 3. This invention combines the dissipative Hamiltonian dynamics model with the symplectic geometric integral algorithm to construct a numerical solution framework for monotonically convergent energy. By introducing a dissipative force term into the dynamic equations and employing an integral step-size update strategy that preserves the symplectic structure, the total system energy is ensured to converge stably to a smaller steady-state value during the iteration process. This guarantees the convergence of the multi-objective optimization problem, enabling the final output control command to accurately track the dispatch commands of the upper-level power grid while satisfying economic cost and network loss optimization. Attached Figure Description

[0019] Figure 1 This is a structural block diagram of the new energy power station operation decision support system of the present invention; Figure 2 This is a flowchart illustrating the operational decision-making method of the present invention.

[0020] Among them, 10 is the multi-source state space reconstruction module; 20 is the Riemannian manifold geometry engine module; 30 is the adaptive inertia Hamiltonian evolution module; and 40 is the symplectic geometric integral and instruction mapping module. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0022] Please see the appendix Figure 1 This invention provides a decision support system for the operation of new energy power stations based on multi-objective optimization. The system mainly includes: a multi-source state space reconstruction module 10, a Riemannian manifold geometry engine module 20, an adaptive inertia Hamiltonian evolution module 30, and a symplectic geometric integral and instruction mapping module 40.

[0023] The multi-source state space reconstruction module 10 is configured to communicate with the underlying data acquisition terminal and the upper-level dispatch center of the new energy power station. The multi-source state space reconstruction module 10 is used to collect the operation data and dispatch instructions of the power station in real time, and map physical quantities into dimensionless generalized coordinate vectors to construct the initial Euclidean state space of the system.

[0024] The Riemannian manifold geometry engine module 20 is connected to the multi-source state space reconstruction module 10. The Riemannian manifold geometry engine module 20 is used to construct the Riemannian metric tensor according to the physical constraints of the site, reconstruct the Euclidean state space into a Riemannian manifold space, and calculate the geometric connection strength factor characterizing the local topological features of the space.

[0025] The adaptive inertia Hamiltonian evolution module 30 is connected to the Riemannian manifold geometry engine module 20. The adaptive inertia Hamiltonian evolution module 30 is used to modulate the virtual inertia matrix of the system in real time according to the geometric connection strength factor, and combine it with the multi-objective potential energy function to construct a dissipative Hamiltonian dynamic model that includes generalized momentum and generalized coordinates.

[0026] The symplectic geometric integral and command mapping module 40 is connected to the adaptive inertia Hamiltonian evolution module 30. The symplectic geometric integral and command mapping module 40 is used to iteratively solve the dynamic model using a numerical integration algorithm with a symplectic structure, obtain the generalized coordinates under steady state, and decode them into control commands for various equipment in the new energy power station.

[0027] Please see the appendix Figure 2 The decision-making process executed by this system includes the following steps: Step S1: The multi-source state space reconstruction module 10 acquires real-time measurement data of the new energy power station at the current moment, including wind turbine output power, photovoltaic inverter output power, energy storage unit state of charge, and node voltage data. Simultaneously, it acquires active and reactive power dispatch instructions issued by the upper-level power grid. The multi-source state space reconstruction module 10 normalizes the above data, defines a decision variable vector including energy storage charging and discharging power, wind and solar curtailment, and reactive power compensation, and establishes a basic potential energy function composed of the weighted sum of each sub-objective function.

[0028] Step S2: The Riemannian manifold geometry engine module 20 receives the decision variable vector and the fundamental potential function. The Riemannian manifold geometry engine module 20 identifies all inequality constraints in the station operation and constructs a Riemannian metric tensor using the second derivative information of the logarithmic barrier function. The Riemannian manifold geometry engine module 20 further calculates the Christofel notation based on the partial derivatives of the metric tensor and uses this to solve for the geometric connection strength factor at the current state point. This curvature scalar is used to quantify the degree of geometric nonlinearity near the constraint boundaries.

[0029] Step S3: The adaptive inertia Hamiltonian evolution module 30 receives the geometric connection strength factor. The adaptive inertia Hamiltonian evolution module 30 calculates the current virtual inertia matrix using a preset curvature and inertia mapping function. As the curvature scalar value increases, the eigenvalues ​​of the virtual inertia matrix increase accordingly. The adaptive inertia Hamiltonian evolution module 30 introduces a generalized momentum variable, combines the basic potential energy function with the modulated virtual kinetic energy term, constructs a generalized Hamiltonian function describing the evolution of the generalized coordinate vector, and establishes a canonical equation of motion including dissipation terms.

[0030] Step S4: The symplectic geometric integral and command mapping module 40 discretizes and solves the canonical motion equations. The symplectic geometric integral and command mapping module 40 uses the symplectic Euler integral algorithm or a higher-order symplectic integral algorithm to update the generalized momentum and generalized coordinates while preserving the symplectic structure characteristics of the system. The symplectic geometric integral and command mapping module 40 monitors the kinetic energy convergence of the system. When the kinetic energy norm is less than a preset threshold, it determines that the system has reached a stable equilibrium point and outputs the optimal generalized coordinates at the current moment.

[0031] Step S5: The symplectic geometric integral and command mapping module 40 restores the optimal generalized coordinates to control parameters with physical dimensions through inverse mapping, including the active power setpoint of the energy storage unit, the active power reduction command of the photovoltaic inverter and the reactive power setpoint, and sends the control parameters to the site coordination controller for execution.

[0032] The multi-source state space reconstruction module 10 is configured to define decision variables and initialize the state space. This module first obtains real-time operating data and external scheduling commands from the renewable energy power plant through a data interface. The input data specifically includes the real-time power of wind and photovoltaic power generation units. The state of charge (SOC) of the energy storage unit and the voltage of each node. and power dispatch reference instructions issued by the superior power grid Based on the physical quantities collected above, the multi-source state space reconstruction module 10 constructs a generalized coordinate vector to describe the operating state of the station. This vector Defined as 3D real space A vector in , where The dimensions of the decision variables. To encompass the active power balance regulation capability and reactive power voltage support capability of the power station, a generalized coordinate vector is used. The specific form is defined as follows: ; in, This represents the charging and discharging power variable of the energy storage unit. The positive and negative values ​​of this variable correspond to the discharge state and charging state of the energy storage system, respectively. It represents the amount of active power curtailment or power regulation of new energy power generation units, used to characterize the degree of active power reduction of photovoltaic or wind turbines based on maximum power point tracking mode; This indicates the reactive power output variable of the reactive power compensation device or inverter in the power station.

[0033] To eliminate the influence of different physical dimensions on the numerical stability of subsequent dynamic evolution calculations, the multi-source state space reconstruction module 10 modifies the generalized coordinate vectors. Each component in the data is dimensionless. This process uses the rated capacity of the power station or a preset benchmark as a reference, mapping measured values ​​with actual physical units (such as kilowatts or volts) to per-unit values ​​or normalized values. After this mapping, the real-time operation problem of the renewable energy power station is transformed into... The problem involves locating geometric points in a 1 / 2 Euclidean space, which serves as the initial state space of the system. Preprocessing operations such as setting the data acquisition frequency, filtering and noise reduction, and communication protocol conversion are well-known techniques to those skilled in the art and will not be elaborated upon here.

[0034] The multi-source state-space reconstruction module 10 constructs the basic potential energy function based on the multiple operational objectives of the new energy power station. Within the framework of generalized Hamiltonian dynamics, this fundamental potential energy function defines the energy distribution pattern in the state space. Its smaller points mathematically correspond to the optimal solution of a multi-objective optimization problem or specific operating points on the Pareto front. The system abstracts objectives such as maximizing economic benefits, achieving accurate grid dispatch command tracking, and minimizing equipment losses into different scalar potential energy components, and uses the principle of minimum energy to guide the evolution of the generalized coordinate vector.

[0035] Fundamental potential function It is constructed using a linear weighted sum to comprehensively represent multiple conflicting operational objectives: ; in, For the first The potential energy component function corresponding to each sub-target This refers to the weighting coefficient corresponding to the sub-objective, used to adjust the relative importance of different control objectives in the decision-making process. To optimize the total number of objectives.

[0036] In specific implementations, in order to simultaneously consider grid connection assessment indicators and the operational economy of the station equipment, the basic potential energy function... It is configured as a composite function containing economic cost potential energy, network loss potential energy, and tracking error potential energy, and its specific form is expressed as: ; In the above formula, This term represents the economic cost potential energy associated with equipment aging, primarily targeting the cycle life losses of energy storage units and the switching losses of photovoltaic inverters. It is constructed based on the power throughput or depth of discharge of the energy storage battery, resulting in higher potential energy values ​​corresponding to high-loss operating conditions. This represents the potential energy term for network losses within the power station, used to characterize the power losses of the collector lines and transformers. This term is positively correlated with the square of the transmission power of each branch. (The third term...) This represents the power point tracking error potential energy term, used to quantify the actual total grid-connected power of the power station. Reference instructions for power dispatch from the upper-level power grid The deviation between them, the square of which constitutes a constraint field similar to elastic potential energy, forces the system output to follow the scheduling instructions. , , These are the weighting coefficients corresponding to the three potential energy terms mentioned above.

[0037] Through the above construction, the system establishes a virtual potential energy field at the mathematical level. When When the deviation from the scheduling instruction target or the resulting excessive economic loss occurs, the corresponding potential energy function value is... This will increase, thus generating a generalized force (i.e., the negative direction of the potential energy gradient) that drives the state vector towards the lower potential energy region in subsequent dynamic evolution. For each sub-function of the potential energy function... The specific parsing format and weighting coefficient values ​​can be configured by those skilled in the art based on actual electricity pricing strategies, equipment parameters, and power grid assessment standards, and will not be elaborated here.

[0038] The Riemannian manifold geometry engine module 20 is used to establish the feasible domain boundary of the operating state of the renewable energy power station. This boundary is defined by a set of operating limit parameters of each physical device within the station. To mathematically represent the physical constraints and embed them into the subsequent manifold geometry structure, the Riemannian manifold geometry engine module 20 first identifies and extracts all inequality constraints at the current moment of the power station, and then uniformly converts these constraints into standard inequality function forms. .

[0039] In practice, the physical constraints handled by the Riemann manifold geometry engine module 20 cover the power and capacity limitations of the energy storage unit, the output limitations of the new energy power generation unit, and the capacity limitations of the inverter.

[0040] For the energy storage unit, the system considers its maximum charge / discharge power limit and the coupling constraint of state of charge on power. The charge / discharge power of the energy storage unit... It must be within the permissible power range Within this range, the power range is not a fixed value, but is dynamically calculated by the battery management system based on the current battery temperature, health status, and real-time SOC. The Riemannian manifold geometry engine module 20 transforms the aforementioned bilateral constraints into two independent one-sided inequality functions: ; For new energy power generation units (photovoltaics or wind turbines), the adjustable range of their active power output is limited by current meteorological resource conditions (such as sunlight intensity and wind speed) and the rated capacity of the inverter. Definition The maximum available renewable energy power at the current moment. This represents the amount of power actively forfeited. The amount of curtailed solar or wind power must be non-negative and cannot exceed the currently available power. The corresponding constraint function is constructed as follows: ; For inverter or converter equipment, the system considers the coupling limitation of apparent power capacity on active and reactive power. The output apparent power of the inverter must not exceed its rated capacity. Due to active power With reactive power A geometric orthogonal relationship exists, and this constraint manifests as a nonlinear circular or elliptical boundary. The Riemannian manifold geometry engine module 20 resolves this nonlinear physical constraint into the following inequality form: ; in, Indicates the maximum available renewable energy power. Indicates the amount of power reduction. This refers to the actual grid-connected active power. This constraint ensures that the decision variables remain within the inverter's safe operation circle during the adjustment process. Through the above analytical process, the Riemannian manifold geometry engine module 20 unifies and abstracts the constraints of discrete and heterogeneous physical devices within the site into a standardized set of analytical inequalities. This set defines the topological manifold regions in the state space that allow the system to operate, providing a clear geometric boundary description for the subsequent construction of the Riemannian metric tensor and curvature calculation. For more complex grid node voltage sensitivity constraints or line thermal stability limits, those skilled in the art can extend them into corresponding inequality functions with reference to the above form, which will not be elaborated further here.

[0041] Based on the aforementioned set of inequality constraint functions, the Riemannian manifold geometry engine module 20 constructs a Riemannian metric tensor to describe the local geometric properties of the state space. To transform the hard safety boundary of a physical system into a geometric structure on a mathematical manifold, this module uses the second derivative information of the logarithmic barrier function to correct the underlying Euclidean metric. This method utilizes the metric transformation principle in Riemannian geometry, changing the length measurement scale of the state space in different regions so that the path length approaching the constraint boundary mathematically tends to infinity. This ensures that the generalized coordinate vector always evolves within the feasible region without using logical truncation or external penalty functions.

[0042] Riemannian metric tensor constructed by Riemannian manifold geometry engine module 20 It is dependent on the state vector. A positive definite symmetric matrix. The analytical expression of this tensor is formed by the weighted sum of the identity matrix and the Hessian matrices of all constraint barrier functions: ; in, represent The identity matrix of dimension corresponds to the flat Euclidean space metric when far from the constraint boundary; This represents the total number of inequality constraints. For the first One constraint function; It is the natural logarithm operator; For generalized coordinate vectors The second-order differential operator (Hesse operator); For the first The scale factor corresponding to each constraint function is used to adjust the range and intensity of the barrier field in space.

[0043] To clarify the specific implementation of this tensor in numerical computation, the Riemannian manifold geometry engine module 20 calculates... Each component element For the state vector of the th Each component and the Each component The formula for calculating the elements of a metric tensor is as follows: ; in, The symbol for Kronecker is used if and only if The value is 1 if it is true, and 0 otherwise.

[0044] The construction of this metric tensor embodies the following technical mechanism: when the generalized coordinate vector When located in the center of the feasible region, the constraint function value Far from zero, the second derivative of the logarithmic barrier term is smaller. Approximately equal to the identity matrix The state space exhibits isotropic and flat characteristics, which is beneficial for fast search in optimization algorithms. When the generalized coordinate vector... Approaching any constrained boundary (i.e.) When ), item The second derivative increases dramatically and approaches positive infinity. This will cause the metric tensor to... The eigenvalues ​​of tend to infinity in this direction. In Riemannian geometry, the square of the infinitesimal distance between two points is defined as... Therefore, small coordinate changes near the boundary It will be a huge metric matrix Magnified to a larger geometric distance This change in geometric properties constitutes a soft constraint mechanism, ensuring that any trajectory attempting to cross the boundary corresponds to an infinitely long path on the Riemannian manifold, thus naturally confining the dynamic evolution process to the feasible region. Furthermore, since the Hessian matrix of the logarithmic function is positive semi-definite under convex constraints, the superposition of the identity matrix guarantees... The positive definiteness over the entire domain satisfies the mathematical existence condition of the Riemannian metric.

[0045] Riemannian manifold geometry engine module 20 further builds upon the Riemannian metric tensor constructed above. This module analytically calculates the geometric connection coefficients and scalar curvature measures that characterize the degree of manifold curvature. To quantify the changes in vector translations in non-Euclidean space, this module first calculates the inverse matrix of the metric tensor. .set up The elements are The inverse matrix satisfies Using the inverse matrix elements of the metric tensor And its first-order partial derivative with respect to generalized coordinates, the Riemannian manifold geometry engine module 20 calculates the second kind of Christofel notation, denoted as This notation describes the Levi-Civita connection on a Riemannian manifold, i.e., the rate of change of the tangent space basis vectors as they move along the manifold. For any three coordinate indices in the state space of dimension , The calculation formula is as follows: ; in, To achieve a dummy index; Represents the metric tensor components Regarding coordinates The partial derivatives. This calculation process extracts the gradient information of the metric field in various directions of space, reflecting the degree of distortion of the geometric structure after embedding constraints in local regions. In order to transform the complex tensor field information into a single variable signal that can be directly used for control and regulation, the Riemannian manifold geometry engine module 20 further calculates the scalar measure of Riemann curvature. Considering that the complete Riemann curvature tensor is a fourth-order tensor, direct calculation involves a large number of tensor shrinking operations, resulting in excessive computational overhead and hindering real-time decision-making. Therefore, this embodiment uses the Christofer-signed Frobenius norm as a proxy variable for curvature intensity to characterize local geometric nonlinearity. Geometric connection strength factor The calculation formula is defined as follows: ; scalar The physical meaning is clear: when the generalized coordinate vector When the metric tensor is in a flat region far from all constraints, Approximating a constant matrix, its partial derivatives are zero, leading to All are zero, at this time This indicates that space exhibits linear characteristics; when the generalized coordinate vector When near the constraint boundary or in a strongly nonlinear region where multiple constraints are intertwined, the metric tensor changes drastically, leading to... The value of increases, thus making The value increases sharply. The Riemannian manifold geometry engine module 20 will calculate the real-time... The values ​​are transmitted to the adaptive inertia Hamiltonian evolution module 30, serving as the core feedback variables for adjusting the system's dynamic characteristics. The specific numerical differentiation algorithms for partial derivatives and the matrix inversion algorithms are standard techniques in computational mathematics and will not be elaborated upon here.

[0046] The adaptive inertia Hamiltonian evolution module 30 is based on the geometric connection strength factor output by the Riemannian manifold geometry engine module 20. Construct a state-dependent virtual inertia matrix This matrix plays a mass role in the generalized Hamiltonian system, determining the response speed and trajectory smoothness of the generalized coordinate vector under the influence of the potential energy gradient. Unlike traditional optimization methods that use fixed step sizes or fixed inertia, this embodiment employs a variable inertia mechanism to address numerical oscillations in non-convex constraint regions and slow convergence in flat regions. The adaptive inertia Hamiltonian evolution module 30 uses the hyperbolic tangent function as the activation function to construct a mapping from curvature scalars to inertia gain. Virtual inertia matrix. The specific parsing expression is defined as follows: ; in, for A real-time virtual inertia matrix of dimensionality; The preset base inertia matrix is ​​usually configured as a diagonal positive definite matrix, with the diagonal elements corresponding to the base time constants or inertial references of each decision variable. This is the upper limit coefficient for inertia gain, used to limit the multiple of the maximum inertia relative to the base inertia, and its value is a positive real number. This is the sensitivity coefficient, used to adjust the response rate of the inertia to changes in curvature; Let be the hyperbolic tangent function, mapping a non-negative curvature scalar to the interval Inside.

[0047] This construction method endows the system with the ability to adaptively adjust to spatial geometric characteristics. When the generalized coordinate vector... When the region is flat within the feasible region, the metric tensor changes little, and the geometric connection strength factor... It approaches zero. At this point, the hyperbolic tangent term... Approaching zero, making Approximately equal to the fundamental inertia matrix In this state, the system maintains a small inertial mass and can accelerate rapidly under the drive of the potential energy gradient, thereby improving the efficiency of convergence to steady state.

[0048] When the generalized coordinate vector When approaching inequality constraint boundaries or in a strongly nonlinear coupling region, the geometric connection strength factor Increase. At this time, The value of increases rapidly and approaches 1, causing the virtual inertia matrix to... Increase to near Physically, this means an increase in the system's virtual mass, leading to greater resistance to changes in motion. This high inertia characteristic effectively suppresses high-frequency numerical oscillations caused by drastic changes in potential energy gradients or repulsive forces from constraint barriers, enhancing the system's trajectory stability near complex constraint boundaries and preventing overshooting or divergence. This can be achieved by adjusting parameters. and Those skilled in the art can flexibly tune the dynamic response characteristics of the system according to the specific requirements for stability and convergence speed in actual application scenarios.

[0049] The adaptive inertia Hamiltonian evolution module 30 constructs a core dynamic model describing the evolution of the generalized coordinate vector. To transform the static optimization problem of new energy power plants into a dynamic physical evolution process, this module focuses on the evolution of the generalized coordinate vector. Based on this, a generalized momentum vector of the same dimension is introduced. Generalized momentum vector belong Space, used to characterize the trend and intensity of changes in the generalized coordinate vector, physically corresponds to the product of virtual mass and generalized velocity. Based on the generalized coordinate vector... and generalized momentum The adaptive inertia Hamiltonian evolution module 30 defines the total energy function of the system, i.e., the Hamiltonian function. The function consists of two parts: a kinetic energy term describing the energy of virtual motion and a potential energy term describing the potential field of the target optimization. Hamiltonian function. The specific mathematical expression is constructed as follows: ; in, This represents the scalar value of the total energy of the system in phase space; This represents the transpose of the generalized momentum vector; This represents the adaptive virtual inertia matrix generated in the preceding steps. The inverse matrix; This represents the fundamental potential function transformed from the multi-objective optimization function. In this Hamiltonian function structure, the first term... This constitutes the generalized kinetic energy on the Riemannian manifold. Due to the virtual inertia matrix... It follows the generalized coordinate vector The changing function (dependent on manifold curvature) means that the kinetic energy term depends not only on the magnitude of the momentum but also on the current spatial geometry. When the system is in a constrained boundary region with high curvature, the inertia matrix... The eigenvalues ​​of increase, leading to the inverse matrix The eigenvalues ​​decrease. This means that, at the same total energy level, the momentum that the system is allowed to hold in complex constrained regions decreases. Or the corresponding speed is suppressed, thus achieving an adaptive adjustment mechanism for deceleration when encountering curves at the physical level. (Second item) This provides the potential field attraction that drives the system toward the optimal solution. By constructing the Hamiltonian function described above, the original non-convex constraint optimization problem is equivalently transformed into an energy conservation or dissipation evolution problem of a physical dynamic system driven by a potential field on a Riemannian manifold.

[0050] The adaptive inertia Hamiltonian evolution module 30 further establishes the dynamic state equations of the system, which describe the generalized coordinate vectors. With generalized momentum vector Over time The evolution law. To ensure that the dynamic trajectory eventually converges to the point of lower potential energy, i.e., the optimal operating solution of the renewable energy station, rather than continuously oscillating on the equipotential surface, this module introduces an energy dissipation mechanism based on the conservative system. Specifically, the system defines a positive definite symmetric dissipation matrix. This is used to simulate the damping effect in a physical system, absorbing excess virtual kinetic energy during the system's evolution. Combined with the previously defined Hamiltonian function... The system's motion follows the generalized dissipative Hamiltonian equations. The specific form of this system of differential equations is: ; in, It represents the derivative of the generalized coordinate vector with respect to time, and its physical meaning is the evolution rate of the system; It represents the derivative of the generalized momentum vector with respect to time, and its physical meaning is the generalized force acting on the system; Represents the Hamiltonian function with respect to the generalized momentum vector The gradient operator; Represents the Hamiltonian function with respect to the generalized coordinate vector. The gradient operator, where It includes the potential energy gradient force and the geometric inertial force caused by the change in the metric tensor field; for A positive definite dissipation matrix of dimension 1 is usually configured as a diagonal matrix in practical implementation, and the values ​​of its diagonal elements determine the damping strength of the system in each control dimension.

[0051] This set of dynamic equations reveals the intrinsic mechanism of system decision generation: the first equation The kinematic constraints of the system are defined, indicating that the state update rate is the product of the momentum and the inverse inertia matrix. This allows the system to operate near the constraint boundary with large curvature (inertia matrix). Increasing the rate of evolution will automatically reduce the rate of evolution. The second equation... The dynamic equilibrium of the system is defined, where The term includes the gravitational force generated by the potential energy gradient (driving the system to move towards a better objective function) and the geometric inertial force generated by the change in the Riemannian metric field. The term then constitutes a dissipative force opposite to the direction of velocity. Under the combined action of the above forces, the total energy of the system decreases monotonically along the trajectory until the system reaches a steady state point where the momentum is zero and the potential energy gradient is in balance with the geometric constraint force.

[0052] The symplectic geometric integral and instruction mapping module 40 is configured to perform numerical discretization of the dynamic equations. Given the special symplectic geometric structure of the Hamiltonian system, to maintain the conservation of phase space volume and ensure the numerical stability of long-term integrals during numerical iteration, this module employs a symplectic-preserving algorithm instead of traditional Runge-Kutta-type algorithms. Specifically, this embodiment selects the first-order symplectic Euler integral method as the core iterative solver. This algorithm, through a semi-implicit update strategy, can effectively suppress numerical energy drift caused by the discretization step size. From time... arrive Within a single iteration cycle, the symplectic geometric integral and instruction mapping module 40 is based on the state at the current moment. Calculate the generalized forces and update the generalized momentum. Then, use the updated generalized momentum... To update generalized coordinates The specific discretization iteration formula is as follows: ; ; in, and Let represent the generalized momentum vectors at the current time and the next time, respectively; and These represent the generalized coordinate vectors at the current and next time points, respectively. The preset time integration step size determines the temporal resolution of the evolution. This represents the partial derivative of the Hamiltonian function with respect to the generalized coordinates calculated at the current state point, i.e., the conservative force term; This represents the dissipation force term based on the current virtual inertia.

[0053] This calculation process demonstrates the staggered update characteristic of the symplectic algorithm: the momentum update explicitly depends on the old position. The update of position implicitly depends on the new momentum. This approach approximately preserves the symplectic form of the continuous Hamiltonian system in the discrete-time domain, making the system more adaptable when the dissipation matrix is ​​introduced. Subsequently, it can strictly follow the law of energy decrease and monotonically converge to the point where the potential energy function is at a smaller value, avoiding the spurious divergence or limit cycle oscillation that may occur in conventional algorithms. The matrix-vector multiplication operations and gradient calculations involved in the iteration process are common algorithm implementations in the field of computational physics, and will not be elaborated here.

[0054] After each iteration, the symplectic geometric integral and instruction mapping module 40 monitors the dynamic convergence state of the system in real time to determine whether the current generalized coordinate vector has evolved to a point with lower energy. To avoid insufficient accuracy or wasted computational resources due to relying solely on iteration count truncation, this module employs an adaptive convergence criterion based on the virtual kinetic energy norm. Specifically, the system calculates the norm of the generalized velocity vector at the current moment if and only if this norm is less than a preset convergence threshold. At this point, the system is considered to have reached steady-state equilibrium. The mathematical expression for the convergence criterion is as follows: ; in, The Euclidean norm operator for vectors; That is, the generalized velocity vector of the system. Physically, it represents the rate of change of decision variables in the state space; It is a preset small positive real number, the specific value of which is set according to the system's requirements for the precision of control commands and the floating-point operation error tolerance of the computing platform.

[0055] Once the above convergence condition is met, the symplectic geometric integral and instruction mapping module 40 locks the generalized coordinate vector at the current moment. As the optimal solution for this decision cycle, the inverse mapping decoding program is initiated. Since the physical quantities were dimensionless during the initialization phase, this module utilizes pre-stored reference parameters (such as the station's rated capacity, rated voltage, etc.) to... Each component is converted into a control command with actual physical units.

[0056] Specifically, the decoded control command set includes the active power setpoint of the energy storage unit. Active power reduction command for new energy inverters and reactive power output command These instructions are precisely transmitted to the corresponding underlying actuators via the industrial control bus within the site. The energy storage battery management system, based on... Adjusting the duty cycle of the IGBTs in the energy storage converter enables precise charge and discharge control; the photovoltaic or wind power inverter controller... and By adjusting its MPPT operating point and reactive current component, precise tracking of grid dispatch commands and optimal allocation of energy flow within the power station can be achieved at the physical level. For message encapsulation and verification techniques during signal transmission, those skilled in the art can implement them using existing communication standards, and will not be elaborated upon here.

Claims

1. A decision support system for the operation of new energy power plants based on multi-objective optimization, characterized in that, include: The multi-source state space reconstruction module (10) collects station operation data and scheduling instructions, and constructs the initial Euclidean state space through the dimensionless mapping method of physical quantities; The Riemannian manifold geometry engine module (20) reconstructs the initial Euclidean state space into a Riemannian manifold space by constructing a Riemannian metric tensor based on physical constraints, and calculates the geometric connection strength factor characterizing local topological features by using the geometric connection coefficient calculation method. The adaptive inertia Hamiltonian evolution module (30) uses the geometric connection strength factor to calculate the virtual inertia matrix through the curvature and inertia mapping function, and combines the multi-objective potential energy function to construct a dynamic model containing dissipation Hamiltonian. The symplectic geometric integral and command mapping module (40) iteratively solves the dissipative Hamiltonian dynamic model using a numerical integration algorithm with a symplectic structure to obtain steady-state generalized coordinates, and decodes the steady-state generalized coordinates into control commands for each device in the new energy power station using an inverse mapping method.

2. The new energy power station operation decision support system based on multi-objective optimization according to claim 1, characterized in that, The components of the generalized coordinate vector constructed by the multi-source state space reconstruction module (10) include at least: a power variable for characterizing the charging and discharging state of the energy storage unit, a curtailment variable for characterizing the degree of active power reduction of the new energy power generation unit, and a reactive power variable for characterizing the reactive power output of the power station reactive compensation device and inverter.

3. The new energy power station operation decision support system based on multi-objective optimization according to claim 1, characterized in that, The multi-objective potential energy function adopts a linear weighted sum form, and the multi-objective potential energy function includes at least: an economic cost potential energy term based on the cycle life loss of energy storage units and the switching loss of inverters, a network loss potential energy term based on the power loss of the internal collection lines and transformers of the power station, and a power tracking error potential energy term based on the deviation between the actual total grid-connected power of the power station and the power dispatch reference command of the upper-level power grid.

4. The new energy power station operation decision support system based on multi-objective optimization according to claim 2, characterized in that, The physical constraints at least include: the charging and discharging power range limit of the energy storage unit, the maximum available power of the new energy power generation unit limited by meteorological resources and the non-negative curtailment of solar and wind power, and the circular boundary limit of the coupling of active and reactive power of the inverter limited by its rated capacity.

5. The new energy power station operation decision support system based on multi-objective optimization according to claim 4, characterized in that, The Riemannian manifold geometry engine module (20) calculates the Riemannian metric tensor by weighted summation of the identity matrix and the second derivative Hesse matrix of the logarithmic barrier function derived from the physical constraints. The Riemann metric tensor is used to make the eigenvalues ​​of the Riemann metric tensor approach infinity when the generalized coordinate vector approaches the constraint boundary.

6. The new energy power station operation decision support system based on multi-objective optimization according to claim 5, characterized in that, The Riemannian manifold geometry engine module (20) calculates the geometric connection strength factor by using the formula for obtaining the Frobenius norm of the second kind of Christofel symbol constructed using the inverse matrix and first-order partial derivative of the Riemannian metric tensor, thereby quantifying the degree of geometric nonlinearity near the constraint boundary.

7. The new energy power station operation decision support system based on multi-objective optimization according to claim 1, characterized in that, The adaptive inertia Hamiltonian evolution module (30) calculates the virtual inertia matrix using a formula for gain modulation of the basic inertia matrix by a hyperbolic tangent function with geometric connection strength factor as the independent variable; the formula for gain modulation is used to smoothly increase the eigenvalues ​​of the virtual inertia matrix when the value of geometric connection strength factor increases.

8. The new energy power station operation decision support system based on multi-objective optimization according to claim 7, characterized in that, The dissipative Hamiltonian dynamics model includes a total energy function and a canonical equation of motion; The total energy function is composed of the sum of the generalized kinetic energy term constructed by the inverse matrix of the virtual inertia matrix and the generalized momentum, and the multi-objective potential energy function; the canonical motion equations include the kinematic equations describing the rate of change of the generalized coordinates and the dynamic equilibrium equations describing the rate of change of the generalized momentum.

9. A new energy power station operation decision support system based on multi-objective optimization according to claim 8, characterized in that, The symplectic geometric integral and instruction mapping module (40) calculates the generalized momentum and generalized coordinates at the next moment through the symplectic Euler integral algorithm, which includes an explicit momentum update step and an implicit coordinate update step involving the Hamiltonian function gradient force and dissipation force; the symplectic Euler integral algorithm is used to maintain the symplectic geometric structure characteristics of the system during the iteration process.

10. The new energy power station operation decision support system based on multi-objective optimization according to claim 1, characterized in that, The symplectic geometric integral and instruction mapping module (40) is used to monitor the generalized velocity vector norm of the system and to determine whether the system has reached a steady state by comparing the generalized velocity vector norm with the convergence threshold. When the steady state is determined, the symplectic geometric integral and instruction mapping module (40) performs inverse mapping decoding and uses the reference value parameters to restore the steady-state generalized coordinates to the active power control command of the energy storage unit, the active power reduction command of the photovoltaic inverter and the reactive power control command.

Images