Shutter type photovoltaic power generation system

Abstract

The invention relates to the technical field of energy management and optimization control of a photovoltaic power generation system, in particular to a shutter type photovoltaic power generation system, which comprises a multi-dimensional state acquisition module, a control module and an energy management module, the disturbance parameter prediction module is used for constructing a short-time domain prediction model and outputting a prediction sequence; the system response and degradation modeling module is used for generating predicted system performance indexes; the system response and degradation modeling module is used for generating a predicted SoH loss amount; the multi-target cost function module is used for constructing a scalar cost function in a prediction time domain; the rolling optimization solving module is used for solving to obtain an optimal control sequence; the control execution and correction module is used for acquiring a new actual state; the control execution and correction module is also used for calculating a prediction error between an actual state and a prediction sequence and carrying out online correction on the disturbance parameter prediction model; according to the method, prospective optimization control is realized, and the prospective performance and effectiveness of decision making are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy management and optimal control of photovoltaic power generation systems, in particular to a louver type photovoltaic power generation system. BACKGROUND

[0002] When the louver type photovoltaic power generation and other building integrated photovoltaic systems are running, multiple and mutually conflicting operation objectives are faced; the system needs to balance between maximizing real-time power generation income, ensuring the long-term health and asset value of energy storage batteries, and maintaining the comfort of indoor light and temperature; Traditional control strategies usually adopt passive response based on preset thresholds or optimization methods for a single target; for example, the control system may only adjust the blades according to the sun position to obtain maximum power generation, or only run to maintain a constant indoor temperature, ignoring the impact of the decision on other target dimensions, especially unable to quantify the long-term cumulative damage of the current operation to the battery life. This way cannot adapt to the dynamic changes of the environment and load, resulting in poor overall operation efficiency of the system; therefore, how to establish an intelligent control method that can prospectively predict future disturbances and quantitatively evaluate and cooperatively optimize the three mutually restrictive targets of power generation, energy storage health, and indoor comfort has become a core technical problem to be solved in the field. SUMMARY

[0003] To solve the above technical problems, the present application provides a louver type photovoltaic power generation system, the technical scheme of the present application comprises: A multi-dimensional state acquisition module for acquiring system operation state data in real time, and constructing a state vector at the current time; A disturbance parameter prediction module for constructing a short-time domain prediction model based on the state vector, and outputting a prediction sequence; A system response and degradation modeling module for establishing a mathematical relationship between control input and key system output, and generating predicted system performance indicators based on the mathematical relationship; the system response and degradation modeling module is also used to construct a nonlinear degradation model, and generate a predicted SoH loss amount based on the nonlinear degradation model; A multi-objective cost function module for constructing a scalar cost function in the prediction time domain by combining the preset weight coefficient, the predicted system performance indicators, and the predicted SoH loss amount; A rolling optimization solving module for solving an optimal control sequence by taking the scalar cost function as the objective function, and combining the state vector, the prediction sequence, and the preset constraint conditions; A control execution and correction module for executing the first control action of the optimal control sequence, and obtaining a new actual state at the next time; the control execution and correction module is also used to calculate a prediction error between the actual state and the prediction sequence, and perform online correction on the disturbance parameter prediction model based on the prediction error.

[0004] Preferably, the state vector comprises: an instantaneous solar irradiance obtained by an irradiance sensor; an ambient temperature, an indoor temperature and a battery pack temperature obtained by a temperature sensor; an indoor illuminance obtained by a photosensitive sensor; a real-time electricity load of the building obtained by a smart meter; and a battery voltage and a battery current obtained by a battery management system.

[0005] Preferably, the disturbance parameter prediction module is specifically configured to: utilize historical solar irradiance and ambient temperature data in the state vector, and output a predicted irradiance sequence by a prediction model based on a time convolution network. The disturbance parameter prediction module is further configured to: utilize historical electricity load and indoor temperature data in the state vector, and output a predicted load sequence by the prediction model. The prediction sequence comprises the predicted irradiance sequence and the predicted load sequence.

[0006] Preferably, the system response and degradation modeling module is specifically configured to: based on the predicted irradiance sequence and a blade angle as a control input, establish a photovoltaic power generation model to generate a predicted power generation amount. The system response and degradation modeling module is further configured to: based on a light transmittance coefficient corresponding to the blade angle and the predicted irradiance sequence, establish an indoor light environment model to generate an indoor illuminance. The system response and degradation modeling module is further configured to: based on a solar heat gain calculation and a first-order lumped parameter thermal model, establish an indoor temperature evolution model to generate an indoor temperature. The predicted system performance indicators comprise the predicted power generation amount, the indoor illuminance and the indoor temperature.

[0007] Preferably, the system response and degradation modeling module is specifically configured to: based on the battery current in the state vector, calculate an instantaneous charge-discharge rate and a discharge depth. The system response and degradation modeling module is further configured to: according to the discharge depth, the battery pack temperature in the state vector, the instantaneous charge-discharge rate, and in combination with a preset degradation rate coefficient, a nonlinear stress index and an apparent activation energy, generate a predicted SoH loss amount.

[0008] Preferably, the multi-objective cost function module is specifically configured to: quantify a deviation degree between the indoor illuminance and the indoor temperature and a preset comfort target value, to obtain an indoor environment deviation penalty term. The multi-objective cost function module is further configured to: combine the predicted power generation amount, the predicted SoH loss amount, the indoor environment deviation penalty term, and a preset power generation revenue weight, a battery health weight and an indoor environment comfort weight, and perform a weighted summation in a prediction time domain, to construct a scalar cost function.

[0009] Preferably, the rolling optimization solving module is specifically used for solving the optimal control sequence under the constraints of the preset safe operation interval of the battery state of charge and the maximum charging and discharging rate, with the goal of minimizing a scalar cost function. The optimal control sequence includes a blade angle sequence and a charging and discharging rate sequence.

[0010] Preferably, the control execution and correction module executes the first control action, specifically the first blade angle control action and the first charging and discharging rate control action in the optimal control sequence. The control execution and correction module calculates the prediction error, specifically the difference between the actual state and the predicted value at the corresponding time in the prediction sequence. The control execution and correction module performs online correction on the disturbance parameter prediction model, specifically updating the model parameters of the disturbance parameter prediction model through the gradient descent algorithm using the prediction error.

[0011] Compared with the prior art, the present application has the following beneficial effects: 1. The system realizes forward-looking optimal control and breaks away from the passive response of the traditional control method to the current state; the system can predict the changes of key disturbance parameters such as solar irradiance and building electricity load in the future short time domain by constructing a prediction model, thereby planning the control strategy in advance and realizing the change from passive response to active prediction, improving the forward-looking and effectiveness of the decision; 2. The system realizes the collaborative optimization of the three conflicting goals of power generation income, battery health and indoor comfort; the system innovatively constructs a unified multi-objective cost function, which quantifies the instantaneous power generation, the predicted battery health loss and the comfort deviation of the indoor environment into a single, solvable cost index, and through the minimization solving of the index, the system can find a dynamic and optimal balance point among multiple mutually restrictive goals; 3. The system converts the fuzzy index of battery long-term health into a quantifiable instantaneous cost, realizing fine management of energy storage assets; the system can accurately calculate the quantified loss of battery life caused by the current control action according to the real-time charging and discharging rate, discharge depth and temperature and other working conditions by constructing a nonlinear degradation model of the battery, which enables the optimization algorithm to actively avoid working conditions that damage the battery life while pursuing short-term power generation income, realizing intelligent trade-off between short-term benefits and long-term asset preservation; 4.The system has closed-loop self-adaptive correction and self-learning ability, ensuring the long-term robustness and high efficiency of the system in real and variable environment; the system adopts a rolling optimization method, re-decides at each time according to the latest actual state, so as to quickly respond to unexpected disturbances, at the same time, the system compares the error between the predicted value and the actual value, and corrects the internal prediction model online, so that it can continuously learn and adapt to the slow changes of seasonal replacement, building characteristics or user habits, and realize the self-evolution of performance. BRIEF DESCRIPTION OF DRAWINGS

[0012] The application will be further explained in connection with the accompanying drawings and embodiments: Figure 1 is a structure diagram of a louver type photovoltaic power generation system. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail in connection with specific embodiments.

[0014] Embodiment 1: Please refer to Figure 1 A louver type photovoltaic power generation system, comprising: A multi-dimensional state acquisition module for acquiring system operation state data in real time and constructing a state vector at the current time; A disturbance parameter prediction module for constructing a short-time domain prediction model based on the state vector and outputting a prediction sequence; A system response and degradation modeling module for establishing a mathematical relationship between control input and key system output, and generating a predicted system performance index based on the mathematical relationship; the system response and degradation modeling module is also used to construct a nonlinear degradation model and generate a predicted SoH loss amount based on the nonlinear degradation model; A multi-objective cost function module for constructing a scalar cost function in the prediction time domain by combining the preset weight coefficient and the predicted system performance index and the predicted SoH loss amount; A rolling optimization solving module for taking the scalar cost function as the objective function and combining the state vector, the prediction sequence and the preset constraint condition to obtain an optimal control sequence; A control execution and correction module for executing the first control action of the optimal control sequence and obtaining a new actual state at the next time; the control execution and correction module is also used to calculate the prediction error between the actual state and the prediction sequence, and correct the disturbance parameter prediction model online based on the prediction error.

[0015] This embodiment provides a louvered photovoltaic power generation system, the purpose of which is to achieve dynamic balance and synergistic optimization among power generation revenue, long-term battery health and indoor environmental comfort through a predictive and adaptive multi-objective optimization control method. In this embodiment, the system includes: The multi-dimensional status acquisition module aims to acquire complete information reflecting the internal and external environment of the system in real time. In this embodiment, it uses a series of sensors such as irradiance sensors, temperature sensors, photosensors, smart meters, and battery management systems (BMS) to collect system operating status data in real time and construct a current-moment status. state vector The state vector is defined as a set that encapsulates all states in the set. The key parameters collected at all times serve to provide unified, high-fidelity real-time data input for subsequent prediction and modeling modules, and are the measurement basis for closed-loop control. The disturbance parameter prediction module aims to anticipate future key environmental changes and load demands, enabling the system to make proactive decisions rather than reactive responses. In this embodiment, it is based on the parameters provided by the multi-dimensional state acquisition module. The historical time series data contained therein are used to construct a short-time domain prediction model. This model is used to output the future. Key prediction sequences within a time step, such as predicted solar irradiance sequences. and predicted building electricity load sequence These predicted sequences will be used as perturbation inputs and invoked by subsequent system response models and cost functions. The system response and degradation modeling module has a dual purpose: first, to establish a forward causal relationship between control inputs and the system's instantaneous performance output; and second, to quantify the cumulative impact of control decisions on the long-term health of the system, especially the battery. In this embodiment, the module is used to establish control inputs such as blade angles. With key system outputs such as power generation , indoor illumination Indoor temperature The mathematical relationship between them, i.e., the forward model of the system response; these models can be based on the predicted inputs from the disturbance parameter prediction module, such as... and the control input to be optimized, such as Generate predicted system performance metrics ; This module is also used to build nonlinear degradation models, particularly for battery health (SoH); these models are based on real-time battery operating conditions such as temperature. Charge / discharge rate This paper uses an empirical aging model to quantify the loss of battery life caused by instantaneous operating conditions, and generates a predicted SoH loss based on this model. ; The multi-objective cost function module aims to unify multiple conflicting operational objectives pursued by the system, such as increased power generation, longer lifespan, and higher comfort, into a single mathematical index that can be solved by an optimization algorithm. In this embodiment, this module incorporates preset weighting coefficients, such as power generation revenue weights. Battery health weight The system response and the predicted system performance indicators output by the degradation modeling module are compared, such as... And the predicted SoH loss as We perform a weighted summation to construct a prediction time domain. Scalar cost function within ; This scalar cost function Defined as an assessment of the future The mathematical expression for the total cost of the step control sequence is used to transform the multi-dimensional performance watts, dimensionless SoH loss, and custom comfort penalty into a unified, dimensionless cost scale through weighting coefficients, making it additive. The rolling optimization solution module aims to calculate the optimal control strategy that minimizes the total future cost while satisfying all safety and physical constraints. In this embodiment, this module uses a scalar cost function constructed by the multi-objective cost function module. The objective function is defined by combining the current state vector obtained from the multi-dimensional state acquisition module and the prediction sequence output by the perturbation parameter prediction module. In addition to a series of preset constraints, such as the battery's safety range, the optimal control sequence is obtained by solving a constrained optimization problem. ;Should The sequence contains the future The optimal blade angle and charge / discharge rate for each step; The control execution and correction module aims to implement the calculated optimal decision and continuously correct the system model through real-time feedback to address uncertainties and model mismatches. In this embodiment, the module employs a rolling time-domain strategy, executing only the optimal control sequence. The first control action in, for example blade angle at time and charge / discharge rate Discard the remaining sequence; in the next time step The module re-obtains the new actual state from the multi-dimensional state acquisition module. ; This module is also used for calculations. The actual state at a given time, for example and Always The predicted sequence at time such as Prediction error between This module is based on prediction error. The model parameters of the perturbation parameter prediction module are obtained through online learning algorithms such as gradient descent. Make online corrections to make its predictions increasingly accurate; This embodiment constructs a complete closed loop of multi-objective predictive control (MOPC) and adaptive correction through the coordinated work of the above six modules; it can not only be based on future predictions and accurate modeling of performance / degradation To solve for the current optimal control action This achieves synergistic optimization of the three major goals of power generation, health, and comfort; moreover, it can utilize the rolling execution of only the first step and closed-loop correction. The two major mechanisms of model updates enable the system to have strong robustness and self-learning ability, continuously adapt to changing environments and user behaviors, and maintain near-optimal energy steady-state operation in uncertain real-world environments.

[0016] Example 2: The state vector includes: instantaneous solar irradiance obtained by an irradiance sensor; ambient temperature, indoor temperature, and battery pack temperature obtained by a temperature sensor; indoor illuminance obtained by a photosensor; real-time building power load obtained by a smart meter; and battery voltage and battery current obtained by a battery management system.

[0017] Based on the implementation method of Example 1, the multi-dimensional state acquisition module in this example, to ensure the comprehensiveness of the state vector, is specifically used for acquisition: The instantaneous solar irradiance obtained by the irradiance sensor is denoted as . ; The ambient temperature obtained by the temperature sensor is denoted as... Indoor temperature, denoted as and battery pack temperature, denoted as ; The indoor illuminance obtained by the photosensor is denoted as . ; The real-time building electricity load obtained through smart meters is denoted as... ; The battery voltage obtained through the battery management system (BMS) is denoted as... And battery current, denoted as ; In this embodiment, the state vector is the vector containing... A set of parameters; This embodiment clearly defines this set of meteorological data, for example... Indoor environment, for example Electrical loads, for example and energy storage status, for example With its multidimensional parameters, this invention ensures the high completeness of the state vector. This complete data input is the high-fidelity foundation for all subsequent advanced function predictions, modeling, and optimizations. The absence of any parameter will lead to inaccurate prediction models or the inability to accurately assess optimization objectives such as comfort and battery health. To ensure the robustness of the system, the multidimensional state acquisition module also incorporates data verification and fault detection logic. This logic filters out obvious outliers by setting reasonable physical boundaries and rate of change limits. When a sensor detects continuous abnormality or loss of data, the system will trigger a fault-tolerant control mode. For example, it may use backup sensor data, use a physical model to estimate the missing data, or switch to a set of preset conservative and safe operating strategies to ensure the basic functions and safety of the system.

[0018] Example 3: Using historical solar irradiance and ambient temperature data from the state vector, a prediction irradiance sequence is output through a prediction model based on a temporal convolutional network. The disturbance parameter prediction module is also used to: utilize historical electricity load and indoor temperature data in the state vector, and output a predicted load sequence through a prediction model; The prediction sequence includes the predicted irradiance sequence and the predicted load sequence.

[0019] Based on the implementation of Example 1, the internal functions of the disturbance parameter prediction module in this example are further specified to achieve high-precision prediction of key disturbances: Predicting Irradiance Sequences: This module utilizes historical solar irradiance data contained in the state vector Xat. and ambient temperature The data, denoted as [data], is predicted using a Temporal Convolutional Network (TCN)-based model. Output the future Predicted irradiance sequence of steps ; Among them, the prediction model based on Temporal Convolutional Network (TCN) Defined as a deep learning model that excels at processing time series data, its function is to capture... and The model is designed to understand the complex time dependencies between different lighting conditions to generate high-precision illumination predictions. Its parameters are obtained through offline training using historical data and are corrected online during operation by the control execution and correction module. Predicting load sequences: This module also utilizes historical electricity load data from the state vector. and indoor temperature data, by another prediction model denoted as , for example, a model such as LSTM or TCN, to output a predicted load sequence for future time steps ; Therefore, the predicted sequence described in embodiment 1, in this embodiment, is explicitly composed of the predicted irradiance sequence and the predicted load sequence ; The specificity and accuracy of the prediction brought by this embodiment; it does not use a single generalized model, but rather a prediction strategy that is divided and conquered, targeting different physical characteristics of perturbed meteorological-related human behavior-related adopted specialized inputs and possibly specialized models such as TCN; this divide-and-conquer prediction strategy significantly improves and the prediction accuracy of the sequence, thus providing a more reliable future scenario input for the subsequent optimization solver, greatly improving the foresight and effectiveness of the MOPC control decision.

[0020] Embodiment 4: Based on the predicted irradiance sequence and the blade angle as a control input, a photovoltaic power generation model is established to generate a predicted power generation; The system response and degradation modeling module is also used to: based on the light transmittance coefficient corresponding to the blade angle and the predicted irradiance sequence, an indoor environment model is established to generate indoor illumination; The system response and degradation modeling module is also used to: based on the solar heat gain calculation and the first-order lumped parameter thermal model, an indoor temperature evolution model is established to generate indoor temperature; Wherein, the predicted system performance indicators include predicted power generation, indoor illumination and indoor temperature.

[0021] On the basis of the implementation manner of embodiment 3, it provides sequence, the system response and degradation modeling module in this embodiment, when constructing the mathematical relationship of the system response, specifically includes the following three core models, which are used to jointly generate the predicted system performance indicators: Photovoltaic power generation model: this model is used to generate predicted power generation ; it is based on the predicted irradiance sequence output by embodiment 3 and the blade angle as a control input , to establish the following mathematical relationship: ; in this formula, representing the predicted power generation unit W, is calculated by this model; representing the photoelectric conversion efficiency dimensionless, its source is the photovoltaic module specification book; representing predicted irradiance in unit of W / m2, is calculated from the sequence of predicted irradiance in Example 3; representing effective photovoltaic area in unit of m2, is derived from photovoltaic module specification; representing dimensionless, is used to characterize the effect of solar incidence angle on photoelectric conversion, is a function of leaf angle and solar elevation angle , and is obtained through optical simulation or experimental calibration; representing leaf angle in unit of degree, is the control variable to be optimized; Indoor light environment model: This model is used to generate indoor illuminance ; it is based on leaf angle corresponding light transmittance and the sequence of predicted irradiance , and the following building optics model is established: ; in this equation, representing predicted indoor illuminance in unit of lm / m2lux, is calculated from this model; representing average air visibility function in unit of lm / W, is used to convert irradiance into illuminance, and is an industry standard constant; representing dimensionless, is a function of leaf angle , and is obtained through experimental calibration or optical simulation; Indoor temperature evolution model: This model is used to generate indoor temperature ; it is a first-order lumped parameter thermal model, which is used to establish the dynamic relationship between thermal gain and indoor temperature ; this model needs to calculate thermal gain through solar heat gain: ; in this equation, representing solar heat gain at time in unit of W; representing window area in unit of m2, is a building design parameter; representing solar heat gain coefficient SHGC in dimensionless, is a function of leaf angle , and is obtained through experimental calibration or thermal simulation; substituting into the indoor temperature evolution model: ; in this evolution model, representing indoor temperature at next time in unit of K or °C, is calculated from this model; representing current indoor temperature, is derived from the sequence of indoor temperature in Example 2; representing time step in unit of s; ​​Represented by the equivalent heat capacity of the building, unit J / K, calibrated by building thermal experiments or simulations; Represented by the unit of heating / cooling capacity of air conditioning and refrigeration, W, as one of the independent control variables to be optimized in the system; Represented by the equivalent heat transfer coefficient of the building shell, unit W / K, calibrated by building thermal experiments or simulations; Represented by the outdoor temperature, which is the source of Example 2 ; This embodiment constructs a forward model or digital twin of the key decision; through these three interrelated physical models, the invention adjusts the blade angle of an abstract control action firmly associated with three specific, quantifiable performance indicators ; this enables the multi-objective cost function module to foresee and quantify the specific impact of any decision on future power generation revenue and indoor comfort, which is the core prerequisite for realizing predictive control.

[0022] Example 5: The system response and degradation modeling module is specifically used to calculate the instantaneous charge / discharge rate and discharge depth based on the battery current in the state vector; The system response and degradation modeling module is also used to generate a predicted SoH loss amount based on the discharge depth, battery pack temperature in the state vector, instantaneous charge / discharge rate, and combined with the preset degradation rate coefficient, nonlinear stress index, and apparent activation energy.

[0023] On the basis of the implementation of Example 2, it provides and The system response and degradation modeling module in this embodiment specifically operates as follows when constructing a nonlinear degradation model: Working condition parameter calculation: this module calculates the key intermediate parameters driving battery aging based on the battery current in the state vector of Example 2: Instantaneous charge / discharge rate value: this parameter is used to quantify the relative size of the current; to match the dimensional requirements of the subsequent empirical degradation model, it is calculated here as a dimensionless value: Where, refers to the rated capacity of the battery, unit Ah, from the battery specification book; is the absolute value of the battery current, unit A; the purpose of this calculation is to convert the actual rate with dimension such as to a dimensionless value such as , so as to be substituted into the subsequent aging model; Depth of discharge : wherein, refers to the current time step Depth of discharge, calculated based on time integration; SoH loss quantification: This module generates the predicted SoH loss amount , the battery pack temperature in the state vector of embodiment 2 , and the instantaneous charge-discharge rate value calculated in the previous step , through a comprehensive SoH degradation rate quantification model, to generate the predicted SoH loss amount ; This model provides an computable function for the optimizer to quantify the long-term health cost of the battery, to punish the working conditions that cause the battery to age too quickly; its mathematical form is as follows: wherein: represents the predicted SoH loss amount, dimensionless, indicating the percentage loss, which is the instantaneous value calculated by this model ; represents the depth of discharge, dimensionless, calculated in the previous step; represents the real-time temperature of the battery in K, from the state vector of embodiment 2 ; represents the dimensionless value of the charge-discharge rate, calculated in the previous step; represents the time step, unit: hour; represents the apparent activation energy of battery aging, unit: J / mol, which is an inherent characteristic parameter of the battery; represents the ideal gas constant, unit: J / mol·K, which is a physical constant; represents the degradation rate coefficient , i.e. , which is an inherent characteristic parameter of the battery; represents the non-linear stress index, dimensionless, which is an inherent characteristic parameter of the battery; To clarify the sources of the above-mentioned battery inherent parameters, they are determined through specific offline calibration experiments; specifically, the calibration process applies multiple groups of different, constant working conditions, for example, at a constant temperature , a constant charge-discharge rate value Constant depth of discharge Accelerated aging tests were conducted under these combined conditions, and the battery cycle life was measured and recorded. Or capacity decay rate data; based on these, which contain multiple sets of operating condition inputs. For example, the corresponding attenuation results The experimental dataset was fitted using regression analysis methods such as nonlinear least squares. The model, thereby identifying This set of inherent characteristic parameters best describes the aging behavior of this type of battery; This embodiment enables the calculability of long-term battery health costs; the model effectively incorporates the complex aging mechanisms at the electrochemical level. The influence is abstracted into a computable mathematical formula, the output of which is... This allows the long-term and vague concept of battery life to be incorporated, for the first time, as a concrete and quantifiable cost item into real-time, short-term optimization objectives. This allows the controller to proactively avoid operating conditions that penalize and damage battery life while pursuing short-term power generation gains, thus achieving an intelligent trade-off between short-term profits and long-term asset preservation. Example 6: The multi-objective cost function module is specifically used to: quantify the degree of deviation between indoor illuminance and indoor temperature and preset comfort target values, and obtain the indoor environment deviation penalty term; The multi-objective cost function module is also used to: combine the predicted power generation, the predicted SoH loss, the indoor environment deviation penalty term, and the preset power generation revenue weight, battery health weight, and indoor environment comfort weight, and perform a weighted summation in the prediction time domain to construct a scalar cost function.

[0024] Based on the implementation methods of Examples 4 and 5, they respectively provide and In this embodiment, the multi-objective cost function module constructs the scalar cost function. The specific operation is as follows: Target and Weight Initialization: This module sets comfort target values, including: expected indoor illuminance. For example, 500 lux and the desired indoor temperature range [ For example, [22°C, 26°C]; the determination of the above comfort target values ​​can be based on industry standards such as the "Design Standard for Indoor Thermal Environment of Buildings", or obtained through user surveys and statistics; at the same time, the weight coefficients of the three major targets are preset: power generation revenue weight. Battery health weight Weighting of indoor environmental comfort ; Comfort Deviation Quantization: The module is based on the predicted indoor illuminance output from Example 4. And predicting indoor temperature Calculate them with the above The degree of deviation from the target value; To solve different physical dimensions, such as and To address the issue that deviations cannot be directly added together, this module employs a normalization penalty method, converting each deviation into a unified, dimensionless penalty term. The penalty is calculated as follows: To further enhance the comprehensiveness of comfort assessment, it is possible to... A glare penalty term is introduced. For example, by calculating the probability of sunlight glare (DGP) and setting a comfort threshold, DGP values ​​exceeding the threshold can be added as an additional penalty term. Furthermore, to more realistically reflect user experience, an asymmetric penalty function can be used instead of the squared term for illumination deviation; for example, for values ​​below the expected value... Illuminance imparts a higher penalty factor; In this formula, represent The indoor environment deviation penalty term at time t is calculated as a dimensionless cost, which is obtained by this model. represent The predicted illuminance and temperature at any given time are from the example of Example 4; This represents the target comfort value, which is initialized and set in step 1 of this process. The penalty factor represents the deviation from photothermal parameters; to ensure Ultimately, it is dimensionless, here These are dimensional tuning parameters, and their function is to adjust the deviation from the square term, which has physical dimensions. and Converting it into a dimensionless denominator makes it additive; for example, The dimensions can be set as follows: , The dimensions can be set as follows: The specific values ​​of these factors can be tuned to balance the relative importance of light and temperature in the overall comfort penalty. Scalar cost function construction: The module combines the predicted power generation output from Example 4. Example 5 output predicted SoH loss and the indoor environment deviation penalty calculated in the previous step. Using the weights preset in step 1 In the prediction time domain The weighted sum is performed over the time horizon, resulting in the final scalar cost function This function is to unify the three conflicting sub-goals into a single scalar cost for optimization; its formulation is as follows: In this function, represents the future step ahead of the current time ; the total cost is dimensionless, and is the objective function of the optimization solver in embodiment 7; represents the prediction horizon length, which is a tuning parameter determined at design time; represents the power generation revenue term, with the negative sign indicating that the system seeks to maximize the power generation revenue, i.e., minimize the negative revenue; represents the battery health cost term; represents the environmental comfort penalty term; represents the weight coefficients; these weights serve as normalization factors to convert different dimensioned physical quantities, such as , which is in units of W, which is dimensionless, etc., into a unified, dimensionless cost scale, making them additive, to ensure the final cost is dimensionless, the weight coefficients need to be dimensionally normalized; for example, the power generation revenue term can be written as , where is a reference power value, in which case is a dimensionless tuning parameter; while and are already dimensionless tuning parameters, by this way, all terms are unified into dimensionless costs, thus ensuring mathematical additivity; This embodiment realizes the quantitative unification of multi-objective decision-making; the cost function is the core of the entire optimization decision; it accurately quantifies the subjective feeling of comfort through the deviation penalty term , and fuses the three completely different dimensional goals of power generation revenue, battery life, and user comfort into a single, minimizable scalar through weighted summation; this fusion not only makes the complex optimization problem solvable, but more importantly, it gives the system extremely high operational flexibility through this set of tuning knobs, allowing users or algorithms to dynamically adjust the strategy preferences according to external conditions such as time-of-use electricity prices, weather, for example, to increase during electricity price peaks, sacrificing comfort in exchange for maximum power generation revenue.

[0025] Embodiment 7: ​The rolling optimization solution module is specifically used to: minimize the scalar cost function and, under the constraints of the preset safe operating range of battery state of charge and the maximum charge-discharge rate, solve for the optimal control sequence. The optimal control sequence includes the blade angle sequence and the charge / discharge rate sequence.

[0026] Based on the implementation method of Example 6, it provides an objective function. In this embodiment, the rolling optimization solution module solves each control cycle. Specifically, it is used to solve a constrained optimization problem: Optimization objective: Minimize the scalar cost function constructed in Example 6. For the goal; Optimization variables: The variables to be solved are the control sequences for the next N steps. ; Constraints: The optimization solution must be performed under preset constraints. These constraints are derived from the initialization settings in step 1 of embodiment 6, and mainly include: Safe operating range for battery state-of-charge SoC: ; Maximum charge / discharge rate: ; Solution process: This module calls an optimization solver, such as a nonlinear programming solver, to solve the following mathematical problem: in Module from Example 2, Modules from Example 3; Solution output: The output calculated by the solver. That is, the optimal control sequence; according to Definition, Explicitly includes the future Step-by-step blade angle sequence and charge / discharge rate sequence and HVAC control sequence ; This embodiment ensures the optimality and security of the decision-making process; if we say If the function defines good criteria, then this module finds the best executor; by solving this constrained optimization problem, this module outputs... It is not just a feasible control sequence, but a comprehensive consideration of the future. Step all prediction information and all goals Afterwards, it can be mathematically guaranteed The optimal sequence with the minimum total cost; simultaneously, the constraints are as follows: The mandatory execution ensures that the system will never exceed the predetermined safety boundary, regardless of the optimization results, thus achieving a balance between high performance and high reliability.

[0027] Example 8: The control execution and correction module executes the first control action, specifically: the first blade angle control action and the first charge / discharge rate control action in the optimal control sequence; The control execution and correction module calculates the prediction error, specifically by calculating the difference between the actual state and the predicted value at the corresponding time in the prediction sequence to obtain the prediction error. The control execution and correction module performs online correction of the disturbance parameter prediction model. Specifically, it updates the model parameters of the disturbance parameter prediction model using the gradient descent algorithm based on the prediction error.

[0028] Based on the implementation method of Example 7, it provides an optimal control sequence. In this embodiment, the control execution and correction module adopts a rolling time-domain strategy combined with adaptive correction, and its operation is specified as three closely linked steps: Execute the first control action: This module does not execute the entire result obtained in Example 7. Instead of sequences, only extract and execute. The first control action in the game: First blade angle control action: ; The first charge / discharge rate control action: ; The first HVAC control action: ; System Execution and and Then, time moves on to the next moment. At this point, the multi-dimensional state acquisition module (see Example 2) will acquire the new actual state. ; Calculate the prediction error: This module utilizes the newly acquired... Actual measured values, such as actual load ,and The time-perturbation parameter prediction module is shown in Example 3. For example, the predicted value at time. By comparing the two, the difference between them is calculated to obtain the prediction error. ; Among them, prediction error The definition of is: Predicted value - Actual value; Online correction of prediction model: The module uses the prediction error calculated in the last step to update the model parameters of the disturbance parameter prediction module, denoted as , such as the weights of the TCN network, through online learning algorithms such as gradient descent SGD; This correction is to improve the accuracy of the prediction model by continuously learning the prediction error; for example, to minimize the mean square error The parameter update law is as follows: For the parameters in this update law: represent the updated model parameter set; represent the current model parameter set, whose dimension is assumed to be dimensionless; represent the prediction error calculated in the last step, whose dimension is consistent with the load , such as watts W; represent the prediction model output unit W, and the gradient of the parameter is dimensionless, so the unit of is also W; To ensure the dimensional consistency of this update law, i.e. the correction term must have the same dimensionless property as itself, and the learning rate is not simply a dimensionless tuning parameter, but a coefficient with a specific physical dimension; In order to make the equation hold, the dimension of must be the load unit , such as ; in this way, the total unit of the entire correction term is: dimensionless This is consistent with the dimensionless property of the model parameter , so that the online correction formula is valid in physical dimension; The technical effect brought by this embodiment is to realize the instant robustness and long-term adaptability of the system; Instant robustness: through the rolling horizon strategy, i.e. only the first step is executed and the next moment the system can immediately respond to unexpected disturbances based on new measurement values , showing high control robustness; Long-term adaptability: through online correction mechanism steps 2 and 3, the system gains the ability of self-learning; prediction error As a feedback signal, continuously drive the parameters of the prediction model Iterative optimization is performed to enable it to automatically track seasonal changes, changes in building physical properties or changes in user habits, thereby continuously reducing prediction error in the long run, ensuring that MOPC decisions are always based on the most accurate predictions, achieving true adaptation and self-evolution.

[0029] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.​

Claims

1. A louver-type photovoltaic power generation system, characterized by ,comprising: a multi-dimensional state acquisition module, configured to acquire system running state data in real time, and construct a state vector at a current time; a disturbance parameter prediction module, configured to construct a short-time domain prediction model based on the state vector, and output a prediction sequence; a system response and degradation modeling module, configured to establish a mathematical relationship between a control input and a key output of the system, and generate a predicted system performance index based on the mathematical relationship; the system response and degradation modeling module is further configured to construct a nonlinear degradation model, and generate a predicted SoH loss amount based on the nonlinear degradation model; a multi-objective cost function module, configured to construct a scalar cost function in a prediction time domain by combining a preset weight coefficient, the predicted system performance index, and the predicted SoH loss amount; a rolling optimization solving module, configured to solve an optimal control sequence by taking the scalar cost function as an objective function, and combining the state vector, the prediction sequence, and a preset constraint condition; a control execution and correction module, configured to execute a first control action of the optimal control sequence, and acquire a new actual state at a next time; the control execution and correction module is further configured to calculate a prediction error between the actual state and the prediction sequence, and correct the disturbance parameter prediction model online based on the prediction error.

2. A louver-type photovoltaic power generation system according to claim 1, characterized in that The state vector comprises: an instantaneous solar irradiance acquired by an irradiance sensor; an ambient temperature, an indoor temperature, and a battery pack temperature acquired by a temperature sensor; an indoor illuminance acquired by a photosensitive sensor; a building real-time power load acquired by a smart meter; and a battery voltage and a battery current acquired by a battery management system.

3. The louvered photovoltaic power system of claim 1, wherein The disturbance parameter prediction module is specifically configured to: utilize historical solar irradiance and ambient temperature data in the state vector, and output a prediction irradiance sequence by a prediction model based on a time convolution network. The disturbance parameter prediction module is further configured to: utilize historical power load and indoor temperature data in the state vector, and output a prediction load sequence by the prediction model. The prediction sequence comprises the prediction irradiance sequence and the prediction load sequence.

4. A louver-type photovoltaic power generation system according to claim 3, characterized in that The system response and degradation modeling module is specifically configured to: establish a photovoltaic power generation model based on the prediction irradiance sequence and a blade angle as the control input, to generate a prediction power generation amount. The system response and degradation modeling module is further configured to: establish an indoor light environment model based on a light transmittance coefficient corresponding to the blade angle and the prediction irradiance sequence, to generate the indoor illuminance. The system response and degradation modeling module is further configured to: establish an indoor temperature evolution model based on solar heat gain calculation and a first-order lumped parameter thermal model, to generate the indoor temperature. The predicted system performance index comprises the prediction power generation amount, the indoor illuminance, and the indoor temperature.

5. The louvered photovoltaic power system of claim 2, wherein The system response and degradation modeling module is specifically configured to: calculate an instantaneous charge-discharge rate and a discharge depth based on the battery current in the state vector. The system response and degradation modeling module is further configured to: generate the predicted SoH loss amount based on the discharge depth, the battery pack temperature in the state vector, the instantaneous charge-discharge rate, and in combination with a preset degradation rate coefficient, a nonlinear stress index, and an apparent activation energy.

6. A louver-type photovoltaic power generation system according to claim 4 or 5, characterized in that The multi-objective cost function module is specifically configured to: quantize the deviation degree between the indoor illuminance and the indoor temperature and the preset comfort degree target value, and obtain an indoor environment deviation penalty term; The multi-objective cost function module is further configured to: combine the predicted power generation, the predicted SoH loss, the indoor environment deviation penalty term, and preset power generation income weight, battery health weight and indoor environment comfort degree weight, and perform weighted summation in the prediction time domain to construct a scalar cost function.

7. A louver-type photovoltaic power generation system according to claim 6, characterized in that The rolling optimization solving module is specifically configured to: solve an optimal control sequence by taking minimization of the scalar cost function as a target and under the constraint of a preset battery state of charge safe operation interval and a maximum charging and discharging rate. The optimal control sequence includes a blade angle sequence and a charging and discharging rate sequence.

8. A louver-type photovoltaic power generation system according to claim 7, characterized in that The control execution and correction module executes a first control action, specifically: executes a first blade angle control action and a first charging and discharging rate control action in the optimal control sequence. The control execution and correction module calculates a prediction error, specifically: calculates a difference value between an actual state and a predicted value at a corresponding moment in the prediction sequence to obtain the prediction error. The control execution and correction module performs online correction on the disturbance parameter prediction model, specifically: updates model parameters of the disturbance parameter prediction model by using the prediction error through a gradient descent algorithm.

Images