A multi-dimensional feature driven light storage power station full life cycle value evaluation system

Abstract

The application belongs to the technical field of photovoltaic storage power station value evaluation, and relates to a multi-dimensional feature driven photovoltaic storage power station whole life cycle value evaluation system, which comprises a multi-dimensional working condition simulation module, a value gradient tensor generation module, a real-time data sensing module, a tensor online calibration module, a target function construction module and a closed-loop instruction execution module.

Description

Technical Field

[0001] This invention belongs to the technical field of photovoltaic-storage power station value assessment, and relates to a multi-dimensional feature-driven full life cycle value assessment system for photovoltaic-storage power stations. Background Technology

[0002] In power systems, the investment and operation of photovoltaic and energy storage power stations require an assessment of the value output of energy storage units throughout their entire life cycle, which serves as the basis for asset allocation and operational strategy formulation. The health degradation of energy storage units is affected by a combination of multiple operating conditions, such as charge / discharge rate, ambient temperature, and cycle depth. Its degradation process exhibits nonlinear and time-varying characteristics, while fluctuations in external electricity market prices further increase the complexity of the life cycle value assessment.

[0003] Chinese Patent Publication No. CN109102185A discloses a method and system for evaluating the economic efficiency of integrated photovoltaic-storage-charging power plants throughout their entire lifecycle. This patent establishes a lifecycle revenue model for integrated photovoltaic-storage-charging power plants, employs a multi-population genetic algorithm to determine the optimal photovoltaic and energy storage capacities for maximizing revenue, and combines a probabilistic model with indicators such as levelized cost of electricity (LCOE) and net present value (NPV) to evaluate economic benefits. This approach primarily focuses on capacity allocation during the project planning phase and macroeconomic calculations based on historical data and static probability distributions.

[0004] However, the aforementioned existing technologies have limitations in practical applications. These methods typically separate economic assessment from the real-time operation control of the underlying equipment, and the static probabilistic models they rely on struggle to accurately capture the instantaneous degradation rate changes of energy storage units caused by multi-dimensional and complex operating conditions in actual operation. Furthermore, static assessment methods are difficult to synchronize deeply with dynamic price signals in the electricity market, making it difficult to directly translate macro-level value assessment results into closed-loop control commands to guide the real-time charging and discharging behavior of the underlying equipment. This can easily cause energy storage assets to deviate from their optimal economic value evolution path during long-term operation.

[0005] Therefore, how to integrate the dynamic physical degradation characteristics of energy storage units under multi-dimensional operating conditions with real-time market economic signals to achieve dynamic assessment and real-time closed-loop optimization control of the full life cycle value of photovoltaic-storage power stations is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a multi-dimensional feature-driven full lifecycle value assessment system for photovoltaic-storage power plants.

[0007] The multi-dimensional operating condition simulation module performs multi-dimensional operating condition simulation, extracts the instantaneous health state attenuation of the energy storage unit in the photovoltaic-storage power station, and synthesizes the benchmark attenuation surface. The value gradient tensor generation module overlays a preset full lifecycle financial model onto a baseline decay surface to settle value loss and constructs an initial state behavior value gradient tensor. The real-time data sensing module acquires short-term market economic signal sequences and internal real-time physical characterization signals of the photovoltaic and energy storage power station, thereby calculating the real-time health status vector. The online tensor calibration module extracts the real decay index from the health state vector to participate in the comparison and prediction deviation error, dynamically corrects the data fixed inside the initial state behavior value gradient tensor, and generates the online updated state behavior value gradient tensor. The objective function construction module calculates electricity revenue based on short-term market economic signal sequences and configures a forward-looking objective function for the value trajectory by combining real-time health status vectors. The optimization control module, based on the value trajectory prospective objective function, frequently calls the online updated state behavior value gradient tensor, and searches and calculates the global objective control instruction sequence along the joint gradient direction of the increasing net value of the objective function; The closed-loop command execution module sends the global target control command sequence to the underlying execution unit of the photovoltaic-storage power station, realizing closed-loop proactive guidance and intervention of the value evolution trajectory.

[0008] A further aspect of the present invention involves synthesizing a reference attenuation surface, comprising the following steps: Deploy an electrochemical-thermodynamic coupling model and import a multi-dimensional enumerated operating condition sequence that includes preset health state values, ambient temperature range, charge / discharge rate range, and cycle depth; An electrochemical-thermodynamic coupling model is applied to perform multidimensional parameter scanning and deduction on a multidimensional enumerated operating condition sequence to calculate the instantaneous health state degradation of the corresponding energy storage unit. The baseline decay surface is synthesized by integrating preset health status values, ambient temperature range, charge / discharge rate range, and the obtained real-time health status decay amount through multivariate function fitting.

[0009] A further aspect of this invention involves constructing an initial state behavior value gradient tensor, comprising the following steps: The impact of the real-time health status decay at each set of coordinate points on the quantitative benchmark decay surface on the reduction of net present value is mapped and output as a scalar value of future value loss. A numerical differential algorithm is applied to solve the first-order partial derivative of the future value loss scalar value with respect to the charging and discharging behavior, thereby obtaining the rate of change of value loss. By organizing and encapsulating environmental parameters, behavioral action dimensions, future value loss scalar values, and the rate of change of value loss, an initial state behavioral value gradient tensor is constructed.

[0010] A further aspect of this invention involves applying a numerical differential algorithm to inversely calculate the first-order partial derivative of the future value loss scalar value with respect to the charging and discharging actions, thereby obtaining the rate of change of value loss. This includes the following steps: Extract the charging and discharging action behavior dimensions corresponding to the current operating condition; Apply a small positive perturbation value to the charging and discharging behavior dimension; The full lifecycle financial model is called again to calculate the scalar value of the value loss after the application of a small disturbance value; Finite difference quotient calculation is performed based on the scalar value of future value loss, the scalar value of value loss after disturbance, and the value of small disturbance. The first-order partial derivative of the output is used as the rate of change of value loss.

[0011] A further aspect of the present invention, which calculates the real-time health status vector, includes the following steps: Read and cache short-term market economic signal sequences that contain information on time-of-use electricity price curves for future periods; The energy storage unit's current operating voltage, loop current, and temperature physical parameters are synchronously captured by sensors to form an internal real-time physical characterization signal. By relying on the online identification algorithm of equivalent circuit parameters based on recursive least squares method to assist in the thermal smoothing of internal real-time physical characterization signals, feature values ​​representing changes in cell internal resistance are extracted, and the true capacity decay index is calculated. Based on this, a real-time health status vector is encapsulated.

[0012] A further aspect of this invention involves generating an online-updated state-behavior value gradient tensor, comprising the following steps: Activate the built-in extended Kalman filter and set the prior inference values ​​recorded in the reference attenuation surface as the reference predicted observation trajectory. The extracted true capacity decay index is introduced as the actual measured observation state parameter, and the evolutionary deviation residual between the basic benchmark predicted observation trajectory and the physical measurement is solved by the extended Kalman filter. Based on the systematic deviation calibration coefficient of the evolutionary deviation residual, the physical benchmark is reconstructed on the benchmark decay surface. Based on the reconstructed benchmark decay surface, the full life cycle financial model is called again to recalculate and correct the partial derivative components of the future value loss scalar value and the rate of change of value loss that are fixed inside the initial state behavior value gradient tensor. The updated state behavior value gradient tensor is then output online.

[0013] A further aspect of this invention involves simultaneously correcting the partial derivative components of the future value loss scalar value and the rate of change of value loss, which are fixed within the initial state behavior value gradient tensor, including the following steps: Based on the generated deviation residual, the inverse derivation is performed by inverting the Kalman gain matrix inside the extended Kalman filter; Based on the inverted derivation results, a set of systematic bias calibration coefficients for the systematic bias of the quantitative model are determined. The calibration coefficients include scaling factors and offsets. By using scaling factors and offsets, the predicted real-time health status decay output of the baseline decay surface is scaled and translated for calibration. The calibrated physical decay data then drives the full life-cycle financial model to solve and overwrite the future value loss scalar value and the partial derivative components of the value loss change rate within the state behavior value gradient tensor.

[0014] A further aspect of this invention involves configuring a forward-looking objective function for the value trajectory, comprising the following steps: We extract price fluctuation factors from short-term market economic signal sequences, deduce and fix mathematical equations for calculating short-term electricity revenues to measure electricity price arbitrage. The predicted basic revenue calculated using the mathematical equation for short-term electricity revenue is subtracted from the scalar value of future value loss at the corresponding stage. Based on this difference, an arithmetic criterion for maximizing net value increment is established. By fixing the real-time health status vector as the physical starting point boundary condition for unfolding evolution calculation, a forward-looking objective function for value trajectory containing spatiotemporal dimensions is constructed.

[0015] A further aspect of this invention involves searching and calculating the global target control command sequence along the joint gradient direction where the net increase in the objective function rises, including the following steps: The gradient controller equipped with the projection gradient maximum ascent optimization engine is activated to determine the physical state and continuous charge and discharge power variables of the current optimization evolution node within the continuous feasible domain of the evolution iteration deduction process. The small changes in the state of charge caused by the continuous charging and discharging power variable within the control time step are mapped into the equivalent damage cycle depth using the rainflow counting algorithm. The equivalent damage cycle depth is then concatenated with the current physical state to form a joint retrieval feature term. The corresponding value loss scalar and the first-order partial derivative with respect to the behavior action are extracted by calling the online updated state behavior value gradient tensor. Substitute the extracted value loss scalar back into the forward objective function of the value trajectory to calculate the current step-by-step net value increase. Combine the partial derivatives of short-term electricity revenue with the extracted first-order partial derivatives to synthesize the joint ascending gradient vector of the objective function. The continuous line search step size extension operation indicated by the joint rising gradient vector is executed in parallel to update the continuous charge and discharge power variable until the increment of the objective function meets the design optimization threshold, and the extreme value control action is locked to generate a global target control command sequence.

[0016] A further aspect of this invention involves distributing a global target control command sequence to the underlying execution unit of the photovoltaic-storage power station to achieve closed-loop proactive guidance and intervention of the value evolution trajectory, including the following steps: The global target control instruction sequence is deconstructed and segmented into microsecond-level timing power control microinstructions adapted to the lower-level hardware. The timing power control micro-instructions are distributed and compressed in parallel to each physical energy storage terminal of the photovoltaic-storage power station via an automated communication bus system for conversion and execution. The heat loss and deformation effects generated after each physical energy storage terminal of the photovoltaic-energy storage power station executes the timing power control micro-instruction are collected. The resulting effects are reset and back-guided to the state monitoring port as a new starting point state constant for updating the next physical operation cycle.

[0017] In summary, the present invention has the following beneficial technical effects: 1. A baseline decay surface incorporating multi-dimensional operating conditions was constructed. By combining an electrochemical-thermodynamic coupling model with multi-dimensional parameter scanning and extrapolation, the interactive effects of ambient temperature, charge / discharge rate, cycle depth, and health status on the instantaneous decay of the energy storage unit were quantified. The decay process was described based on physical mechanisms, reducing the model errors introduced by the traditional linear depreciation assumption with a fixed number of cycles when dealing with complex operating conditions, and providing objective physical evolution data support for subsequent value assessment.

[0018] 2. By overlaying a full life-cycle financial model onto a baseline decay surface and performing numerical differentiation, the physical decay caused by a single charge-discharge cycle is transformed into a scalar of monetized reduction in the future total net present value and its gradient vector with respect to control variables, and encapsulated as a state-behavior value gradient tensor. This establishes a differentiable quantifiable channel from underlying electrochemical behavior to top-level economic value, enabling the acquisition of the long-term economic costs of candidate control behaviors during optimization decisions, thereby providing economic guidance signals for operational control.

[0019] 3. An extended Kalman filter is used to compare the actual capacity decay index calculated online with the predicted trajectory in the value gradient tensor using residuals, and to calibrate the loss scalar and gradient weights in the tensor in reverse. This online update mechanism can dynamically correct systematic evaluation biases caused by individual device differences or long-term changes in the operating environment, alleviate the problem of decoupling between the offline model and the physical entity over long-term operation, and help maintain the adaptability of the control model throughout the entire life cycle of the energy storage unit.

[0020] 4. The real-time health status vector, dynamic market economic signals, and online-updated value gradient tensor are uniformly integrated into the value trajectory look-ahead objective function. This is combined with an optimization controller to generate a global target control command sequence for execution, while simultaneously transmitting the resulting effects of execution back to update the starting state. This mechanism achieves a closed-loop flow from multi-dimensional state perception and dynamic value assessment to control command execution. It effectively integrates the dynamic decay characteristics of energy storage units with real-time market price signals into closed-loop control, guiding the actual operating trajectory of the photovoltaic-storage power station towards a path with higher comprehensive value throughout its entire lifecycle. This solves the problem of the separation between value assessment and underlying real-time control in traditional methods. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention.

[0022] Figure 1 This is a schematic diagram of the evaluation system framework provided in the embodiments of this application.

[0023] Figure 2 This is a schematic diagram of the evaluation method provided in the embodiments of this application.

[0024] Figure 3 This is a schematic diagram of the stacked reference attenuation surface provided in the embodiments of this application.

[0025] Figure 4 This is a multidimensional feature reference attenuation curve provided in the embodiments of this application.

[0026] Figure 5 This is a graph illustrating the principle of solving the first-order partial derivative of value loss using the finite difference method provided in this application embodiment.

[0027] Figure 6 This is a bar chart of global target control instruction sequence execution data provided in the embodiments of this application. Detailed Implementation

[0028] The following is in conjunction with the appendix Figure 1 - Figure 6 A preferred description of the present invention is provided below.

[0029] See attached document Figure 1 - Figure 2 This invention proposes a multi-dimensional feature-driven full life-cycle value assessment system for photovoltaic-storage power stations, comprising the following modules: The multi-dimensional operating condition simulation module performs multi-dimensional operating condition simulation, extracts the instantaneous health state attenuation of the energy storage unit in the photovoltaic-storage power station, and synthesizes the benchmark attenuation surface. The value gradient tensor generation module overlays a preset full lifecycle financial model onto a baseline decay surface to settle value loss and constructs an initial state behavior value gradient tensor. The real-time data sensing module acquires short-term market economic signal sequences and internal real-time physical characterization signals of the photovoltaic and energy storage power station, thereby calculating the real-time health status vector. The online tensor calibration module extracts the real decay index from the health state vector to participate in the comparison and prediction deviation error, dynamically corrects the data fixed inside the initial state behavior value gradient tensor, and generates the online updated state behavior value gradient tensor. The objective function construction module calculates electricity revenue based on short-term market economic signal sequences and configures a forward-looking objective function for the value trajectory by combining real-time health status vectors. The optimization control module, based on the value trajectory prospective objective function, frequently calls the online updated state behavior value gradient tensor, and searches and calculates the global objective control instruction sequence along the joint gradient direction of the increasing net value of the objective function; The closed-loop command execution module sends the global target control command sequence to the underlying execution unit of the photovoltaic-storage power station, realizing closed-loop proactive guidance and intervention of the value evolution trajectory.

[0030] In one embodiment of the present invention, the multi-dimensional working condition simulation module is used to perform the following steps: Deploy an electrochemical-thermodynamic coupling model and import a multi-dimensional enumerated operating condition sequence that includes preset health state values, ambient temperature range, charge / discharge rate range, and cycle depth; An electrochemical-thermodynamic coupling model is applied to perform multidimensional parameter scanning and deduction on a multidimensional enumerated operating condition sequence to calculate the instantaneous health state degradation of the corresponding energy storage unit. The baseline decay surface is synthesized by integrating preset health status values, ambient temperature range, charge / discharge rate range, and the obtained real-time health status decay amount through multivariate function fitting.

[0031] To build a data foundation that reflects the physical degradation characteristics of energy storage units under different operating conditions, an electrochemical-thermodynamic coupled model is deployed on a server equipped with high-performance computing resources. This model simulates the electrochemical reactions and heat generation and conduction processes within the energy storage unit. Specifically, the Doyle-Fuller-Newman model can be used, coupled with a lumped-parameter thermal model to describe the battery's temperature changes.

[0032] In engineering implementation, the chemical-thermodynamic coupling model adopts the P2D model based on pseudo-two-dimensional theory as the core electrochemical framework, and integrates the heat balance equation to take into account the heat generation and heat dissipation effects during the charging and discharging process. Finally, the model is built and solved using COMSOL Multiphysics simulation software or open-source professional libraries such as PyBaMM.

[0033] The system defines and generates a multi-dimensional enumerated operating condition sequence. This sequence is a set composed of multiple parameter dimensions, used to systematically define simulation boundaries and cover typical operating scenarios experienced by the energy storage unit throughout its entire lifecycle. The multi-dimensional enumerated operating condition sequence consists of parameters in four dimensions, namely preset health state values. Ambient temperature range Charge / discharge rate range and loop depth .

[0034] Preset health status value Defined as the ratio of current usable capacity to rated capacity, its value is empirically set to gradually decrease from 1.0 to 0.8, with a step size of 0.01. This is based on the industry consensus that when the capacity of a lithium-ion battery decays to 80% of its initial capacity, that is... SOH When the resistance is 0.8, it reaches the end of its energy storage application life cycle. At this time, the battery internal resistance increases significantly and there is a risk of thermal runaway. Therefore, 0.8 is taken as the lower limit.

[0035] Ambient temperature span symbol The temperature range is set between -20℃ and 45℃, with a step size of 5℃. This range is based on the fact that it fully covers the extreme ambient temperatures of most outdoor photovoltaic and energy storage power stations in the world throughout the four seasons. If the temperature exceeds this range, the battery management system will usually trigger temperature control protection or cut off the circuit.

[0036] Charge / discharge rate range Defined as the ratio of charging / discharging current to rated capacity, its value range is set between 0.1C and 2.0C, with a step size of 0.1C. This is based on the fact that 0.1C to 0.5C covers conventional energy-type application scenarios such as peak shaving and valley filling, while 1.0C to 2.0C covers high-frequency power-type application scenarios such as primary frequency regulation of the power grid.

[0037] Loop depth Defined as the percentage of electricity discharged in a single cycle relative to the initial electricity, with a value range of 0.1 to 1.0 and a step size of 0.1, it is based on the fact that it covers the discharge depth boundaries that may be induced by all grid dispatch commands from shallow charging to full charging and discharging.

[0038] The system programmatically generates all Cartesian product combinations of these parameters, forming a discrete set of simulation task inputs. A deterministic simulation is then performed on each combination of operating conditions in the multidimensional enumerated sequence of operating conditions using multidimensional parameter scanning derivation.

[0039] In each simulation task, the electrochemical-thermodynamic coupled model, based on the input initial health state, temperature, rate capability, and cycle depth, calculates the capacity loss of the energy storage unit after one cycle by solving partial differential equations related to the lithium-ion concentration distribution, potential distribution, and SEI film growth at the solid electrolyte interface. This capacity loss is quantified as the energy storage unit's instantaneous health state decay. This quantifies the decline in the health status of the energy storage unit after completing one full charge-discharge cycle under target operating conditions. Specifically, the instantaneous health status degradation of the energy storage unit is calculated using the following formula:

[0040] In the formula, This represents the available capacity at the start of the current cycle, and its value is determined by... With battery rated capacity The product is determined; This represents the available capacity after one cycle, calculated through simulation using an electrochemical-thermodynamic coupling model. This refers to the rated capacity specified by the energy storage unit at the time of manufacture.

[0041] Once the simulation calculations for all operating condition combinations are completed, the system integrates the input four-dimensional operating condition parameters with the simulation output of the energy storage unit's instantaneous health status degradation. By employing multivariate function fitting techniques, such as radial basis function interpolation or training a surrogate model neural network, discrete data points are synthesized into a continuous high-dimensional mathematical expression, generating a benchmark degradation surface that can query the corresponding degradation amount based on any operating condition input.

[0042] The baseline decay surface is a high-dimensional data structure that can be implemented as a multidimensional lookup table or a trained surrogate model, establishing a model based on the input conditions. To output The mapping relationship is established. This baseline decay surface is stored in the form of a data structure or function model as input for subsequent steps. Assuming the application scenario is the simulation of lithium iron phosphate batteries, whose decay mechanism is mainly based on SEI film growth, the above parameter range is a typical setting in this field.

[0043] In this embodiment, it is assumed that the object being processed in the current step is the rated capacity. This is a 100Ah lithium iron phosphate energy storage unit. The system selects one operating condition combination from a multi-dimensional enumerated operating condition sequence for simulation. The parameters of this combination are: preset health state values. The value is 0.95, and the ambient temperature is... At 25℃, charge / discharge rate 1.0C, loop depth The value is 0.8. Calculate the initial available capacity of the loop. .

[0044] The system invokes the deployed electrochemical-thermodynamic coupled model, inputs operating parameters, and executes a single-cycle simulation involving discharge at 100A to a DOD of 0.8 followed by full charge at 100A. After simulation calculation, the model outputs the available capacity after this cycle. The energy storage capacity is 94.997 Ah. The system calculates the instantaneous health status degradation of the energy storage unit based on the formula. The system generates a data point whose content is ( ). , , , , This process iterates through all combinations in the multidimensional enumeration sequence of operating conditions and uses all the generated data points to construct the final reference attenuation surface.

[0045] Considering that the baseline decay surface integrates multiple dimensions such as ambient temperature range, charge / discharge rate range, and cycle depth, it is essentially a high-dimensional tensor space. To visually demonstrate the nonlinear interaction between these four-dimensional features and physical decay on a drawing, please refer to [link to relevant documentation]. Figure 3 When analyzing this multidimensional map, it is necessary to clarify that... Figure 3 The curved surface entity itself does not represent the "cycle depth". The spatial shape and height (i.e., the Z-axis) of the curved surface represent the amount of instantaneous health status decay.

[0046] Since a conventional three-dimensional coordinate system can only continuously display a maximum of three variables, this embodiment employs a dimensionality reduction visualization method of "multi-layer discrete mapping slices" to overcome the dimensionality limitation and introduce a fourth input parameter, "loop depth." Specifically, Figure 3 In the diagram, the X-axis represents the ambient temperature range, the Y-axis represents the charge / discharge rate range, and the Z-axis represents the instantaneous health status decay. Based on this three-dimensional physical space, and with a fixed preset health status value of 0.95, the "cycle depth" is used as a discrete fourth-dimensional variable. Four independent physical mapping surfaces are calculated and plotted in the same frame when the cycle depths are 0.4, 0.6, 0.8, and 1.0, respectively.

[0047] A systematic comparison of these four surfaces reveals the coupling effect of multidimensional variables. As the cycle depth parameter gradually increases, the surface transitions sequentially from the bottom blue layer (cycle depth 0.4) to the top red layer (cycle depth 1.0). During this process, not only does the overall baseline value of the instantaneous health state decay increase significantly, but the "wrinkling" gradient of the surface becomes more pronounced at the boundaries of high-rate and extreme ambient temperature conditions. This four-level dynamic evolution map reveals that "cycle depth," as an independent fourth dimension, has a significant nonlinear acceleration effect on the instantaneous decay of energy storage units, thus overcoming the limitation of conventional single-layer surfaces in failing to reflect multidimensional characteristics. This provides a high-dimensional physical data foundation for subsequently establishing the state-behavior value gradient tensor.

[0048] See appendix Figure 4 This figure shows the two-dimensional cross-sectional mapping of the baseline degradation surface under different conditions. The figure illustrates four typical operating condition combinations using different line types. The first solid line represents the ideal baseline degradation curve under an ambient temperature range of 25℃ and a preset health state value of 1.0; the second dashed line represents the high-temperature condition with an ambient temperature range of 45℃; the third dotted line represents the low-temperature condition with an ambient temperature range of -20℃. It can be seen that compared to the 25℃ baseline, extreme temperatures significantly increase the overall degradation; the fourth dotted line represents the baseline degradation per cycle shifting upwards due to battery aging when the preset health state value decays to 0.8 at 25℃. Figure 4 It demonstrates the interactive effects and differentiated mapping relationships between changes in ambient temperature range, charge / discharge rate range, and preset health state value on the instantaneous health state decay, providing physical input for the subsequent generation of the value gradient tensor.

[0049] In one embodiment of the present invention, the value gradient tensor generation module is configured to perform the following steps: The impact of the real-time health status decay at each set of coordinate points on the quantitative benchmark decay surface on the reduction of net present value is mapped and output as a scalar value of future value loss. A numerical differential algorithm is applied to solve the first-order partial derivative of the future value loss scalar value with respect to the charging and discharging behavior, thereby obtaining the rate of change of value loss. By organizing and encapsulating environmental parameters, behavioral action dimensions, future value loss scalar values, and the rate of change of value loss, an initial state behavioral value gradient tensor is constructed.

[0050] Specifically, the value gradient tensor generation module maps physical-level decay to economic-level value loss and calculates its sensitivity to control actions. The system loads a pre-built full lifecycle financial model from local storage or a remote server. The full lifecycle financial model internally includes initial equipment investment, equipment residual value discount curves, rolling electricity price forecast curves for the next 20 years, and long-term operation and maintenance cost models.

[0051] The system applies a full lifecycle financial model to the data base of the benchmark decay surface. For each operating point defined on the benchmark decay surface and its corresponding instantaneous health state decay of the energy storage unit, the full lifecycle financial model is used to calculate and quantify the impact of a single cycle of decay on the total net present value of the energy storage unit over its entire lifecycle. Specifically, the system calculates the reduction in the total future arbitrageable electricity or the number of ancillary services that can be provided due to this decay under the current health state, and converts this amount into monetary value using an electricity price and cost model, i.e., a continuous scalar value of future value loss. Is with state and behavior The relevant scalar represents the execution action. The net loss of future total economic value caused by the immediate decay. Among these, the state... Representatives such as and A set of state variables, behavior Representatives such as and A set of controllable variables. The scalar value of this future value loss. The calculation is performed using the following formula:

[0052] In the formula, the function Used to calculate given an initial health state The total net present value that an energy storage unit can create over its remaining lifespan.

[0053] To determine the impact of controlled actions on value loss, the system applies a numerical differentiation algorithm to each operating point on the baseline decay surface. By applying small perturbations to charging / discharging actions, such as charge / discharge rate or cycle depth, and then re-invoking the aforementioned value calculation process, the system reverse-engineers the first-order partial derivative of the future value loss scalar value with respect to that action, thereby obtaining the rate of change of value loss. (Value loss rate of change) It is a vector that relates to the action. The gradient is expressed as:

[0054] The components of this vector are behavior The partial derivatives of each component, for example and The gradient vector is approximated using the finite difference method, i.e., in the behavior... Add a tiny step-size perturbation in one of its dimensions. Recalculate And solve for the difference quotient. and Currency units such as yuan, and All are dimensionless relative values; gradient If the behavior The component in is the charge / discharge rate. The basic dimension is In engineering, "C" is often used as the unit, so the derivative components... The dimension of this can be expressed as yuan / C, and its physical meaning is the increase in future value loss caused by each 1C increase in the charge / discharge rate, i.e., 1 unit rate, under the current state.

[0055] The system encapsulates the environmental parameters, behavioral action dimensions, future value loss scalar value, and value loss change rate associated with this working point into a structured data entity, namely the initial state-behavior-value gradient tensor. The value gradient tensor index is the state. and behavior The discretized combination stores a data pair, containing the corresponding data for that point. Values ​​and gradient vectors The system stores this tensor in a cache or in-memory database for later use.

[0056] In this embodiment, the operating points and their output results, i.e., the states, in the optimization control module example are further processed. for ,Behavior for Real-time health status degradation of energy storage units The system loads a full lifecycle financial model, assuming the solution yields... When the value is 0.95, the remaining total net present value of the energy storage unit It is 15,000 yuan. And when Become At that time, its remaining total net present value It becomes 14998.8 yuan. The scalar value of future value loss under this working condition is calculated. In behavior of Apply small step-size perturbations in the dimension ,make .

[0057] System query in status and new behaviors The attenuation amount is assumed to be The financial model is used again to calculate the new future value loss. The gradient components are calculated using the finite difference method. The system packages the calculation results into an entry in the initial state-behavior-value gradient tensor, with the index being... The value is .

[0058] See appendix Figure 5 The horizontal axis represents the charging and discharging behavior dimension, which is expressed here as the charging and discharging rate. For example, the vertical axis represents the scalar value of future value loss, and the solid line in the figure is the full life-cycle value depreciation curve, showing that as the charge / discharge rate increases, the economic depreciation caused by the physical degradation of the energy storage unit exhibits a non-linear upward trend. The figure marks two key calculation points: before the disturbance, when the charge / discharge rate is 1.0C, the corresponding future value loss scalar value is 1.2 yuan; after applying a small positive disturbance of 0.1C, the charge / discharge rate reaches 1.1C, and the value loss scalar value calculated by the re-applied full life-cycle financial model rises to 1.8 yuan. The dashed line in the figure represents the secant line connecting these two calculation points, and its slope is the first-order partial derivative obtained by the finite difference method, i.e., the rate of change of value loss, with a value of 6 yuan / C. This partial derivative quantifies the marginal economic penalty brought about by each unit increase in the charge / discharge rate under the current operating conditions.

[0059] In one embodiment of the present invention, the real-time data sensing module is configured to perform the following steps: Read and cache short-term market economic signal sequences that contain information on time-of-use electricity price curves for future periods; The energy storage unit's current operating voltage, loop current, and temperature physical parameters are synchronously captured by sensors to form an internal real-time physical characterization signal. By relying on the online identification algorithm of equivalent circuit parameters based on recursive least squares method to assist in the thermal smoothing of internal real-time physical characterization signals, feature values ​​representing changes in cell internal resistance are extracted, and the true capacity decay index is calculated. Based on this, a real-time health status vector is encapsulated.

[0060] Specifically, the real-time data sensing module acquires the real-time status of the power plant. The system first establishes a communication connection with the external power grid dispatch data center through a data acquisition interface program, based on the IEC61850 standard or the Modbus / TCP protocol. The interface program actively requests and reads the time-of-use electricity price curves and bidding and demand information for ancillary services such as frequency regulation and reserve services for the next 24 to 72 hours at a preset time frequency. This information is temporarily stored in a circular buffer, forming a short-term market economic signal sequence. The short-term market economic signal sequence is a dynamically updated dataset with a key-value pair array, where the key is a timestamp and the value is the corresponding predicted electricity price or service bid.

[0061] Simultaneously, sensor network hardware deployed within the photovoltaic-storage power station, including voltage sensors, Hall effect current sensors, and temperature sensors, synchronously acquires data. An industrial control computer or data acquisition unit (RTU) continuously obtains the current terminal voltage, loop current, and surface temperature physical parameters of the energy storage unit via a bus. These physical quantities are packaged into timestamped, synchronized data frames, constituting an internal real-time physical characterization signal. This signal is essentially a time-series data stream containing at least three channels, such as voltage, current, and temperature.

[0062] The system's health status calculation module processes the acquired real-time internal physical characterization signals. This module executes an online equivalent circuit parameter identification algorithm based on recursive least squares to identify the polarization resistance and ohmic resistance of the energy storage unit in real time. The health status calculation module is also supplemented by thermal smoothing logic, utilizing the acquired surface temperature... The internal temperature of the battery was estimated using a simplified equivalent thermal resistance model. This value is used to correct the thermistor parameter in the internal resistance-capacity mapping model, and finally the true capacity decay index is calculated. The actual capacity degradation index is a scalar value or a set of scalar values ​​that represents the degree of decrease in the actual usable capacity of an energy storage unit relative to its factory-rated value.

[0063] True capacity decay index Other real-time physical states are encapsulated into a real-time health state vector, where other real-time physical states are a one-dimensional array including the current state of charge. Real-time terminal voltage Real-time loop current and surface temperature Specifically, it is expressed as .

[0064] In the embodiment, the system at time Perform this step. The system receives time-of-use electricity price data for the next 24 hours, forming a short-term market economic signal sequence, such as {"t_0+1h": 0.8 yuan / kWh,"t_0+2h": 1.2 yuan / kWh,...}. Simultaneously, the sensor acquisition terminal voltage is 3.2V, the loop current is 50A, and the surface temperature is 30℃, which are packaged to form the internal real-time physical characterization signal [3.2,50,30].

[0065] Over the past complete cycle, the total charge input was calculated to be 98 Ah, and the total charge output to be 97 Ah. The thermal smoothing unit compensated for the coulombic efficiency based on temperature, calculating the actual usable capacity to be 97 Ah. (This is for the rated capacity.) Given 100Ah, the actual capacity decay index of the energy storage unit was calculated. The value is 0.97. The current state of charge is estimated based on the voltage and OCV curves. The value is 0.65. The system integrates and encapsulates the data into a real-time health status vector. This information is then passed to subsequent steps for use.

[0066] In one embodiment of the present invention, the tensor online calibration module is configured to perform the following steps: Activate the built-in extended Kalman filter and set the prior inference values ​​recorded in the reference attenuation surface as the reference predicted observation trajectory. The extracted true capacity decay index is introduced as the actual measured observation state parameter, and the evolutionary deviation residual between the basic benchmark predicted observation trajectory and the physical measurement is solved by the extended Kalman filter. Based on the systematic deviation calibration coefficient of the evolutionary deviation residual, the physical benchmark is reconstructed on the benchmark decay surface. Based on the reconstructed benchmark decay surface, the full life cycle financial model is called again to recalculate and correct the partial derivative components of the future value loss scalar value and the rate of change of value loss that are fixed inside the initial state behavior value gradient tensor. The updated state behavior value gradient tensor is then output online.

[0067] Specifically, the online tensor calibration module calibrates and corrects the initial state behavior value gradient tensor. The system activates the built-in extended Kalman filter, whose state transition equation and observation equation describe the model of value loss evolution over time. The filter's state space model is pre-configured to correspond to the structure of the initial state behavior value gradient tensor. The system sets the time series predictions of the future value loss scalar value and the rate of change of value loss, which match the current environment and operating conditions in the initial state behavior value gradient tensor, as the baseline prediction observation trajectory. This trajectory refers to the sequence generated by the filter based solely on state predictions from its internal model when there is no external measured data input.

[0068] The system inputs the actual capacity decay index of the energy storage unit extracted from the real-time data sensing module as the actual measured state parameter into the extended Kalman filter. This decay index, i.e., the actual SOH value, constitutes the true observed value of the system's evolution trajectory. The extended Kalman filter executes a prediction-update loop, quantifying the difference between the predicted observation trajectory and the physically measured evolution by solving the state equation and the observation equation, thus forming the evolutionary deviation residual. Is the filter in Actual measurement value at time Compared with predicted measurements The difference between them:

[0069] In the formula, It is a nonlinear observation equation function; For the system in The prior state estimate at time step 1 represents the model prediction error. The evolutionary deviation residual is obtained. Then, the system is reverse-engineered using the Kalman gain matrix to calculate the systematic deviation calibration coefficients, including scaling factors, etc. and offset It should be noted that here... and Instead of preset static constants, these are based on the covariance matrix of the Kalman filter and the current evolutionary deviation residuals. The compensation weights obtained through dynamic iteration are used to adaptively absorb the systematic biases of the model caused by sudden changes in battery aging mechanisms or temperature sensor drift.

[0070] The system utilizes systematic deviation calibration coefficients to reconstruct the physical baseline of the reference attenuation surface. The predicted instantaneous health state attenuation is output from the reference attenuation surface. Perform scaling and translation calibration, and correct the formula:

[0071] In the formula To correct for the true physical attenuation, the system will Substituting the data back into the full lifecycle financial model, a nonlinear financial mapping is performed to solve and overwrite the scalar value of future value loss and the partial derivative components of the rate of change of value loss within the state-behavior value gradient tensor. The corrected data structure is the online-updated state-behavior value gradient tensor, which is stored in memory and reflects the actual loss characteristics of the current energy storage unit.

[0072] In the embodiment, assume the current time The extended Kalman filter then begins to operate. The filter predicts the next time step. Energy storage unit health status The value is 0.9695, forming part of the baseline predicted observation trajectory. The real-time data sensing module outputs the time... Actual measured capacity degradation index of energy storage units The value is 0.9700, which is used as the input filter for the actual measured observation state parameters. The resulting deviation residual is then calculated. Positive residuals indicate that the model prediction overestimated the degradation rate of the energy storage unit. The filter is back-derived to update the internal state, outputting a systematic bias calibration coefficient, assuming a scaling factor. Offset Assuming the future value loss scalar value for a certain operating condition is 1.2 yuan, and corrected using a calibration coefficient, and assuming the original predicted attenuation value output by the reference surface is... The value was 0.000038. After correction, Similarly, the weights of each component of the gradient vector are adjusted. The corrected values ​​together constitute the updated state-behavior value gradient tensor.

[0073] In one embodiment of the present invention, the objective function construction module is configured to perform the following steps: We extract price fluctuation factors from short-term market economic signal sequences, deduce and fix mathematical equations for calculating short-term electricity revenues to measure electricity price arbitrage. The predicted basic revenue calculated using the mathematical equation for short-term electricity revenue is subtracted from the scalar value of future value loss at the corresponding stage. Based on this difference, an arithmetic criterion for maximizing net value increment is established. By fixing the real-time health status vector as the physical starting point boundary condition for unfolding evolution calculation, a forward-looking objective function for value trajectory containing spatiotemporal dimensions is constructed.

[0074] Specifically, the objective function construction module constructs a forward-looking objective function for the value trajectory. This module analyzes the short-term market economic signal sequence acquired from the real-time data perception module, extracts the time-of-use electricity prices for various future time segments, and derives the short-term electricity revenue mathematical equation. The electricity revenue mathematical equation takes charging / discharging power, duration, real-time electricity price, and charging / discharging efficiency as inputs, and outputs the monetary value of the basic revenue. Specifically, the short-term electricity revenue mathematical equation... It is a piecewise function used to calculate at time t. With power The formula for the immediate benefits or costs generated by charging and discharging operations is as follows:

[0075] In the formula, To control the power, i.e. the DC side charging and discharging power of the energy storage unit, a value greater than 0 indicates discharging, and a value less than 0 indicates charging; The moment obtained from the short-term market economic signal sequence Electricity price; and These are the discharge and charging efficiencies of the energy storage unit, respectively. It is the time step of control.

[0076] Based on the forward optimization objective, the global optimal objective is set as maximizing net value appreciation. The system calculates the predicted basic revenue using the aforementioned short-term electricity revenue mathematical equation, and subtracts the future value depreciation scalar obtained by querying the online-updated state-behavior value gradient tensor. This difference is defined as the net increase in value of a single decision within a time step.

[0077] The system uses the real-time health state vector obtained by the real-time data sensing module, including the current true SOH and SOC, as the physical starting point boundary condition for the optimization calculation. This physical starting point boundary condition corresponds to the initial state in the objective function. By integrating boundary conditions, the arithmetic criterion for maximizing net value increment, and the revenue and cost model, a forward-looking objective function for the value trajectory is constructed. It is a method for optimizing the time domain Control power sequence within Let be the cumulative function of the independent variable, where For the first Control power at each time step For the first The predicted electricity price for each time step is expressed as follows:

[0078] In this embodiment, it is assumed that the scheduling for the next hour needs to be optimized, with a time step of... Optimize the time domain to 15 minutes. The electricity price for the first 15 minutes can be determined from short-term market economic signal sequences. The price is 0.8 yuan / kWh. Real-time health status vector. Set as physical starting point boundary conditions Evaluate candidate actions as the first time step. Discharge, assuming discharge efficiency The base return is 0.95, and the predicted base return is... .

[0079] by state and candidate actions For the index, query the state-behavior-value gradient tensor after online updates, assuming a calibrated future value depreciation scalar. It is 1.5 yuan. The net increase per step is... The objective function is to maximize the sum of the net increases at these four time steps. .

[0080] In one embodiment of the present invention, the optimization control module is configured to perform the following steps: The gradient controller equipped with the projection gradient maximum ascent optimization engine is activated to determine the physical state and continuous charge and discharge power variables of the current optimization evolution node within the continuous feasible domain of the evolution iteration deduction process. The small changes in the state of charge caused by the continuous charging and discharging power variable within the control time step are mapped into the equivalent damage cycle depth using the rainflow counting algorithm. The equivalent damage cycle depth is then concatenated with the current physical state to form a joint retrieval feature term. The corresponding value loss scalar and the first-order partial derivative with respect to the behavior action are extracted by calling the online updated state behavior value gradient tensor. Substitute the extracted value loss scalar back into the forward objective function of the value trajectory to calculate the current step-by-step net increase. Combine the partial derivatives of short-term electricity revenue with the extracted first-order partial derivatives to synthesize the joint ascending gradient vector of the objective function.

[0081] The optimization control module seeks the optimal control strategy for the prospective objective function of the value trajectory. The system launches a multi-step gradient controller on the main control server, which includes a projected gradient steepest ascent optimization engine. The projected gradient optimization module is used to find the maximum value of the prospective objective function of the value trajectory within the continuous power variable space. Within a preset evaluation period, the controller iterates at discrete time steps. At each iteration step, the algorithm initializes a set of continuous charge and discharge power variables within the hardware-allowed upper and lower power limits (continuous feasible region).

[0082] To address the physical scale difference between transient control step size and periodic cycle depth of DOD, the controller extracts the continuous charge / discharge power variable within a single control time step. The continuous charging and discharging trajectory caused by the internal flow is used to extract the amplitude and mean of the local charging and discharging cycle using the rainflow counting algorithm, and then mapped to the equivalent damage cycle depth within that time period. The controller will then determine the physical state of the current node (e.g., ... SOH The surface temperature is concatenated with the equivalent damage cycle depth and power variables to form a joint retrieval feature term. Using this feature term as a key, the controller queries the online updated state-behavior value gradient tensor, extracting the corresponding current step value depreciation scalar (FVL) value and its first-order partial derivative with respect to the action. .

[0083] After acquiring the data, the controller performs the following calculations in parallel: On the one hand, it substitutes the extracted value loss scalar back into the forward objective function of the value trajectory to calculate the projected net increase within the time slice. On the other hand, calculate the partial derivative of short-term electricity revenue with respect to power. And, combined with the partial derivatives of value loss, synthesize the joint ascending gradient vector of the objective function. This joint gradient vector effectively combines the economic incentive of earning electricity in the short term with the economic penalty of damaging equipment value in the long term.

[0084] Subsequently, the steepest ascent optimization engine utilizes the synthesized joint ascent gradient vector to perform a line search step-size extension operation within the continuous feasible region, i.e., along... The direction of the fastest increase in the indicated net increment moves the search point a continuous distance determined by the Hessian matrix approximation or the step size factor, thereby updating the continuous charge and discharge power variable. When the increment of the objective function in two consecutive iterations is less than a preset design optimum threshold (e.g., ...), ... The convergence process terminates when the maximum number of iterations is reached. This threshold is set to... The basis for this is that in the millisecond-level dispatch response scenario of photovoltaic and energy storage power stations, this accuracy can ensure that the economic value assessment error of power commands is controlled within the minimum legal tender settlement accuracy, i.e. Within the order of magnitude, it fully meets the lossless requirements of actual power transaction settlement. Further reducing the threshold would lead to excessively long traversal time for the fastest ascent optimization engine, resulting in adverse consequences such as delayed control command issuance. Therefore, this threshold is a balance point that takes into account both solution accuracy and real-time response of the underlying communication. The system locks the continuous power control actions that exhibit extreme values ​​throughout the entire iteration process and generates a global target control command sequence covering the evaluation cycle.

[0085] In an embodiment, the controller is in the first... Each time step iterates. The optimization engine initializes a continuous power variable at the current evolution node, let's say it's... With a control time step of 15 minutes and a rated capacity of 100 kWh, the controller calculates that the power transfer caused by this power within 15 minutes is 12.5 kWh, resulting in an equivalent damage cycle depth of 0.125.

[0086] The controller concatenates the physical state with (power 50kW, equivalent damage cycle depth = 0.125) and queries the updated state-behavior value gradient tensor. It returns the value loss scalar. Yuan, and the first-order partial derivative of loss discount Yuan / kW. Simultaneously, the short-term revenue partial derivative is calculated based on the real-time electricity price. Yuan / kW. This leads to the synthesis of a joint ascending gradient vector. Yuan / kW.

[0087] Since the joint gradient is positive, it indicates that the electricity cost benefit from increasing power outweighs the discounted cost of equipment losses. The steepest ascent optimization engine searches for a step size based on this positive gradient extension line, updating the evaluated power in the next internal iteration. This process is repeated until the gradient approaches zero (marginal gain equals marginal loss) or the increment of the objective function is minimal. The engine converges and establishes the optimal continuous power for that time step (assuming convergence to 85.3kW), and combines the time steps into a global target control command sequence such as {t1: 85.3kW, t2: -60.1kW, t3: 92.5kW, ...} and outputs it.

[0088] See appendix Figure 6 The graph illustrates the distribution of the output control command sequence along the time axis. Red bars represent discharge control commands with power > 0, such as 85.3kW at step t1 and 92.5kW at step t3. Green bars represent charging control commands with power < 0, such as -60.1kW at step t2 and -45.0kW at step t4. This graph reflects the system's decision evolution, driven by the joint gradient ascent engine, through high-frequency switching of charge and discharge states, achieving a balance between maximizing short-term power gains and minimizing long-term battery losses.

[0089] In one embodiment of the present invention, the closed-loop instruction execution module is configured to perform the following steps: The global target control instruction sequence is deconstructed and segmented into microsecond-level timing power control microinstructions adapted to the lower-level hardware. The timing power control micro-instructions are distributed and compressed in parallel to each physical energy storage terminal of the photovoltaic-storage power station via an automated communication bus system for conversion and execution. The heat loss and deformation effects generated after each physical energy storage terminal of the photovoltaic-energy storage power station executes the timing power control micro-instruction are collected. The resulting effects are reset and back-guided to the state monitoring port as a new starting point state constant for updating the next physical operation cycle.

[0090] Specifically, the closed-loop instruction execution module implements closed-loop control of the power plant operation. The system's internal instruction parsing and distribution module deconstructs the global target control instruction sequence, dividing it into timing-based power control micro-instructions measured in microseconds or milliseconds. These micro-instructions are binary data frames or messages conforming to the underlying hardware communication protocol and control logic. This process converts macroscopic power instructions into specific pulse-width modulation (PWM) duty cycles, current setpoints, or voltage setpoints that are directly recognizable by the lower-level machine.

[0091] The system uses a power station-level automated communication bus system to send timing power control micro-instructions in parallel to the execution nodes of each physical energy storage terminal in the photovoltaic-storage power station, namely the lower-level inverter PCS responsible for AC / DC conversion and the battery management system BMS responsible for battery safety management. After receiving the micro-instructions, these execution nodes execute them and adjust the output power to match the instruction requirements.

[0092] After executing the command, the system uses sensors deployed inside and on the surface of the device to collect in real time the derivative effects generated by the execution nodes of the physical energy storage terminal after executing the micro-command. These effects include heat loss due to Joule heating, deformation of the material's microstructure, and other microscopic causes of energy storage unit degradation. The system quantifies the derivative effect data and feeds it back to the state monitoring port located in the cloud or edge server. This port collects data from sensors and models, using the feedback data as update parameters to correct and set the initial state constants for the next optimization iteration.

[0093] Through a cyclical interlocking effect of command issuance, execution, status feedback, and model updates, the system achieves a closed-loop control mechanism for the physical power plant. This cyclical interlocking effect includes a short-range real-time control closed loop and a long-range strategic optimization closed loop. The interlocking of these two loops ensures that the system can respond to instantaneous changes while evolving towards the goal of maximizing long-term value.

[0094] In this embodiment, the system generates a global target control command sequence in the previous module, with a command power of 85.3kW. The command parsing module generates timing power control micro-instructions: setting the maximum allowable discharge current in the BMS Controller Area Network (CAN) message; and setting the active power setpoint to 85.3kW in the PCS Modbus / TCP message. The micro-instructions are sent to the BMS and PCS. After receiving the instructions, the PCS adjusts the PWM wave of its IGBT transistor switches, causing the energy storage unit's DC side to discharge at a power of 85.3kW.

[0095] During discharge, the BMS sensor detected a temperature rise from 30.1℃ to 30.3℃, which is the derivative effect. This 0.2℃ temperature rise data is fed back to the status monitoring port. Before the next optimization cycle, when the system updates the real-time health status vector, 30.3℃ is used as the new state constant. The next baseline decay surface query and future value depreciation scalar calculation are both based on this updated state, thus achieving closed-loop control.

[0096] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0097] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A multi-dimensional feature-driven full life-cycle value assessment system for photovoltaic-storage power stations, characterized in that, include: The multi-dimensional operating condition simulation module performs multi-dimensional operating condition simulation, extracts the instantaneous health state attenuation of the energy storage unit in the photovoltaic-storage power station, and synthesizes the benchmark attenuation surface. The value gradient tensor generation module overlays a preset full lifecycle financial model onto a baseline decay surface to settle value loss and constructs an initial state behavior value gradient tensor. The real-time data sensing module acquires short-term market economic signal sequences and internal real-time physical characterization signals of the photovoltaic and energy storage power station, thereby calculating the real-time health status vector. The online tensor calibration module extracts the real decay index from the health state vector to participate in the comparison and prediction deviation error, dynamically corrects the data fixed inside the initial state behavior value gradient tensor, and generates the online updated state behavior value gradient tensor. The objective function construction module calculates electricity revenue based on short-term market economic signal sequences and configures a forward-looking objective function for the value trajectory by combining real-time health status vectors. The optimization control module, based on the value trajectory prospective objective function, frequently calls the online updated state behavior value gradient tensor, and searches and calculates the global objective control instruction sequence along the joint gradient direction of the increasing net value of the objective function; The closed-loop command execution module sends the global target control command sequence to the underlying execution unit of the photovoltaic-storage power station, realizing closed-loop proactive guidance and intervention of the value evolution trajectory.

2. The multi-dimensional feature-driven full life-cycle value assessment system for photovoltaic-storage power stations according to claim 1, characterized in that, The process of synthesizing the reference attenuation surface includes the following steps: Deploy an electrochemical-thermodynamic coupling model and import a multi-dimensional enumerated operating condition sequence that includes preset health state values, ambient temperature range, charge / discharge rate range, and cycle depth; An electrochemical-thermodynamic coupling model is applied to perform multidimensional parameter scanning and deduction on a multidimensional enumerated operating condition sequence to calculate the instantaneous health state degradation of the corresponding energy storage unit. The baseline decay surface is synthesized by integrating preset health status values, ambient temperature range, charge / discharge rate range, and the obtained real-time health status decay amount through multivariate function fitting.

3. The multi-dimensional feature-driven full life-cycle value assessment system for photovoltaic-storage power stations according to claim 1, characterized in that, Constructing the gradient tensor of the initial state behavior value includes the following steps: The impact of the real-time health status decay at each set of coordinate points on the quantitative benchmark decay surface on the reduction of net present value is mapped and output as a scalar value of future value loss. A numerical differential algorithm is applied to solve the first-order partial derivative of the future value loss scalar value with respect to the charging and discharging behavior, thereby obtaining the rate of change of value loss. By organizing and encapsulating environmental parameters, behavioral action dimensions, future value loss scalar values, and the rate of change of value loss, an initial state behavioral value gradient tensor is constructed.

4. The multi-dimensional feature-driven full life-cycle value assessment system for photovoltaic-storage power stations according to claim 3, characterized in that, The numerical differentiation algorithm is used to inversely solve for the first-order partial derivative of the future value loss scalar value with respect to the charging and discharging behavior, thereby obtaining the rate of change of value loss. This includes the following steps: Extract the charging and discharging action behavior dimensions corresponding to the current operating condition; Apply a small positive perturbation value to the charging and discharging behavior dimension; The full lifecycle financial model is called again to calculate the scalar value of the value loss after the application of a small disturbance value; Finite difference quotient calculation is performed based on the scalar value of future value loss, the scalar value of value loss after disturbance, and the value of small disturbance. The first-order partial derivative of the output is used as the rate of change of value loss.

5. The multi-dimensional feature-driven full life-cycle value assessment system for photovoltaic-storage power stations according to claim 1, characterized in that, The real-time health status vector is calculated by the following steps: Read and cache short-term market economic signal sequences that contain information on time-of-use electricity price curves for future periods; The energy storage unit's current operating voltage, loop current, and temperature physical parameters are synchronously captured by sensors to form an internal real-time physical characterization signal. By relying on the online identification algorithm of equivalent circuit parameters based on recursive least squares method to assist in the thermal smoothing of internal real-time physical characterization signals, feature values ​​representing changes in cell internal resistance are extracted, and the true capacity decay index is calculated. Based on this, a real-time health status vector is encapsulated.

6. The multi-dimensional feature-driven full life-cycle value assessment system for photovoltaic-storage power stations according to claim 5, characterized in that, Generating the updated state-behavior-value gradient tensor online includes the following steps: Activate the built-in extended Kalman filter and set the prior inference values ​​recorded in the reference attenuation surface as the reference predicted observation trajectory. The extracted true capacity decay index is introduced as the actual measured observation state parameter, and the evolutionary deviation residual between the basic benchmark predicted observation trajectory and the physical measurement is solved by the extended Kalman filter. Based on the systematic deviation calibration coefficient of the evolutionary deviation residual, the physical benchmark is reconstructed on the benchmark decay surface. Based on the reconstructed benchmark decay surface, the full life cycle financial model is called again to recalculate and correct the partial derivative components of the future value loss scalar value and the rate of change of value loss that are fixed inside the initial state behavior value gradient tensor. The updated state behavior value gradient tensor is then output online.

7. The multi-dimensional feature-driven full life-cycle value assessment system for photovoltaic-storage power stations according to claim 6, characterized in that, The partial derivative components of the future value loss scalar value and the rate of change of value loss, which are fixed within the initial state behavior value gradient tensor, are corrected synchronously, including the following steps: Based on the generated deviation residual, the inverse derivation is performed by inverting the Kalman gain matrix inside the extended Kalman filter; Based on the inverted derivation results, a set of systematic bias calibration coefficients for the systematic bias of the quantitative model are determined. The calibration coefficients include scaling factors and offsets. By using scaling factors and offsets, the predicted real-time health status decay output of the baseline decay surface is scaled and translated for calibration. The calibrated physical decay data then drives the full life-cycle financial model to solve and overwrite the future value loss scalar value and the partial derivative components of the value loss change rate within the state behavior value gradient tensor.

8. The multi-dimensional feature-driven full life-cycle value assessment system for photovoltaic-storage power stations according to claim 1, characterized in that, Configure the forward objective function for the value trajectory, including the following steps: We extract price fluctuation factors from short-term market economic signal sequences, deduce and fix mathematical equations for calculating short-term electricity revenues to measure electricity price arbitrage. The predicted basic revenue calculated using the mathematical equation for short-term electricity revenue is subtracted from the scalar value of future value loss at the corresponding stage. Based on this difference, an arithmetic criterion for maximizing net value appreciation is established. The real-time health status vector is fixed as the physical starting point boundary condition for the unfolded evolution calculation, and a value trajectory prospective objective function containing spatiotemporal dimensions is constructed by fusing and constructing it.

9. The multi-dimensional feature-driven full life-cycle value assessment system for photovoltaic-storage power stations according to claim 1, characterized in that, The global objective control command sequence is calculated by searching along the joint gradient direction that leads to the increase in the net value of the objective function, including the following steps: The gradient controller equipped with the projection gradient maximum ascent optimization engine is activated to determine the physical state and continuous charge and discharge power variables of the current optimization evolution node within the continuous feasible domain of the evolution iteration deduction process. The small changes in the state of charge caused by the continuous charging and discharging power variable within the control time step are mapped into the equivalent damage cycle depth using the rainflow counting algorithm. The equivalent damage cycle depth is then concatenated with the current physical state to form a joint retrieval feature term. The corresponding value loss scalar and the first-order partial derivative with respect to the behavior action are extracted by calling the online updated state behavior value gradient tensor. Substitute the extracted value loss scalar back into the forward objective function of the value trajectory to calculate the current step-by-step net value increase. Combine the partial derivatives of short-term electricity revenue with the extracted first-order partial derivatives to synthesize the joint ascending gradient vector of the objective function. The continuous line search step size extension operation indicated by the joint rising gradient vector is executed in parallel to update the continuous charge and discharge power variable until the increment of the objective function meets the design optimization threshold, and the extreme value control action is locked to generate a global target control command sequence.

10. The multi-dimensional feature-driven full life-cycle value assessment system for photovoltaic-storage power stations according to claim 1, characterized in that, The global target control command sequence is sent to the underlying execution unit of the photovoltaic-storage power station to achieve closed-loop proactive guidance and intervention of the value evolution trajectory, including the following steps: The global target control instruction sequence is deconstructed and segmented into microsecond-level timing power control microinstructions adapted to the lower-level hardware. The timing power control micro-instructions are distributed and compressed in parallel to each physical energy storage terminal of the photovoltaic-storage power station via an automated communication bus system for conversion and execution. The heat loss and deformation effects generated after each physical energy storage terminal of the photovoltaic-energy storage power station executes the timing power control micro-instruction are collected. The resulting effects are reset and back-guided to the state monitoring port as a new starting point state constant for updating the next physical operation cycle.

Images