Distributed photovoltaic power station dynamic pricing method, device, equipment and storage medium

Abstract

The embodiment of the present application relates to the field of artificial intelligence and new energy technology, and discloses a kind of distributed photovoltaic power station dynamic pricing method, device, equipment and storage medium, the method comprises: comprehensive health index determination step, the comprehensive health index of photovoltaic power station is determined based on multi-source heterogeneous data;Residual service life prediction step, obtain the health state sequence of photovoltaic power station equipment history, according to health state sequence and preset sequence prediction model, predict the residual service life of equipment;Pricing model construction step, according to the comprehensive health index and residual service life, a dynamic pricing model is constructed;Pricing step, predict the future failure probability of photovoltaic power station, determine the future maintenance cost of photovoltaic power station according to future failure probability, determine the pricing of photovoltaic power station according to future maintenance cost and market correction value in pricing model.The above-mentioned manner, the embodiment of the present application can improve the accuracy and efficiency of photovoltaic power station pricing.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and new energy technologies, specifically to a dynamic pricing method, device, equipment, and storage medium for distributed photovoltaic power plants. Background Technology

[0002] With the acceleration of energy transition, distributed photovoltaic (PV) power plants, as an important component of clean energy, have experienced rapid development. Currently, the installed capacity of distributed PV exceeds 200GW, forming a huge existing asset market. However, the following technical issues exist in the pricing of PV power plants:

[0003] Existing pricing methods are static, mainly relying on human experience, simple annual depreciation methods, capacity unit price methods, or income present value methods. Because these methods use a single indicator or a single model for pricing, they cannot accurately quantify the actual health status of the equipment, and they ignore multi-dimensional factors such as the nonlinear degradation characteristics of photovoltaic modules, the complex aging mechanism of inverters, and the cumulative impact of environmental factors on equipment lifespan. Therefore, the pricing model is singular, and the accuracy and efficiency of the pricing results are low. Summary of the Invention

[0004] In view of the above problems, embodiments of the present invention provide a dynamic pricing method, apparatus, equipment and storage medium for distributed photovoltaic power plants to solve the problems existing in the prior art.

[0005] According to one aspect of the present invention, a dynamic pricing method for distributed photovoltaic power plants is provided, the method comprising:

[0006] The comprehensive health index determination steps involve collecting multi-source heterogeneous data from distributed photovoltaic power plants, and determining the comprehensive health index of the photovoltaic power plants based on the multi-source heterogeneous data. The multi-source heterogeneous data includes at least the equipment data, operation monitoring data, environmental data, historical maintenance data, performance evaluation data, and market data of the photovoltaic power plants.

[0007] The remaining useful life prediction step involves obtaining the historical health status sequence of the photovoltaic power station equipment, and predicting the remaining useful life of the equipment based on the health status sequence and a preset sequence prediction model.

[0008] The step of constructing a pricing model involves building a dynamic pricing model based on the comprehensive health index and the remaining lifespan. This step includes:

[0009] The basic value of the photovoltaic power station is determined based on the remaining useful life.

[0010] The basic value is adjusted based on the comprehensive health index to obtain the health-adjusted value;

[0011] The market dynamic factors of the photovoltaic power station are obtained, and the market is adjusted based on the market dynamic factors and the health adjustment value to obtain the market adjustment value.

[0012] The pricing process involves predicting the future failure probability of the photovoltaic power station, determining the future maintenance cost of the photovoltaic power station based on the future failure probability, and determining the price of the photovoltaic power station based on the future maintenance cost and the market correction value in the pricing model.

[0013] In one alternative approach, the comprehensive health index determination step includes:

[0014] Collect multi-source heterogeneous data from distributed photovoltaic power plants, and determine the component health index, inverter health index, system health index, and reliability index of the photovoltaic power plants based on the multi-source heterogeneous data and a preset deep learning model. The deep learning model is trained using a multi-task learning strategy.

[0015] The weight combination is automatically learned based on the training of the deep learning model. The comprehensive health index H is then determined based on the weight combination, the component health index, the inverter health index, the system health index, and the reliability index.

[0016] ,

[0017] H1 is the component health index, H2 is the inverter health index, H3 is the system health index, and H4 is the reliability index. The weights in the weight combination include w1, w2, w3, and w4, and w1+w2+w3+w4=1.

[0018] In one alternative approach, the sequence prediction model is a temporal convolutional network model, which includes multiple concatenated blocks. Each block contains residual connections and multiple causal dilated convolutional layers with different dilation rates. The multiple causal dilated convolutional layers simultaneously capture the temporal features of the health state sequence at multiple time scales from short-term to long-term. The outputs of the multiple causal dilated convolutional layers are fused together as the output features of the block.

[0019] In one alternative approach, the temporal convolutional network model is trained using a loss function that incorporates physical laws, wherein the loss function L is: Where L_pred is the prediction error, L_physics is the physical constraint term, and γ is the weight parameter. The physical constraint term is constructed based on the physical degradation model of the device, and the performance degradation of the device follows a Weibull distribution.

[0020] In one alternative approach, determining the fundamental value of the photovoltaic power station based on the remaining useful life includes:

[0021] Based on the remaining useful life, the design life of the photovoltaic power station and the life value function learned by the deep learning model, the basic value is determined based on the life value function and the market price of a newly built photovoltaic power station of the same scale.

[0022] Among them, the basic value V_new is the market price of the newly built photovoltaic power station of the same scale, f_life is the lifetime value function, R is the remaining lifetime, and L_design is the design lifetime of the photovoltaic power station.

[0023] , where a is the fitting parameter;

[0024] The step of adjusting the baseline value based on the comprehensive health index to obtain a health-adjusted value includes: determining a health adjustment function based on the comprehensive health index and its variance; determining the health-adjusted value based on the health adjustment function and the baseline value; and specifying the health-adjusted value as the adjusted value. ,

[0025] Where g_health is the health correction function, H is the comprehensive health index, and σ is the variance of the comprehensive health index;

[0026] The market dynamic factors include market supply and demand balance factors and technology iteration depreciation factors. The market supply and demand balance factors are predicted based on a preset time-series network model. The process of obtaining the market dynamic factors of the photovoltaic power station and performing market corrections based on the market dynamic factors and the health correction value to obtain the market correction value includes: performing market corrections based on the market supply and demand balance factors, the technology iteration depreciation factor, and the health correction value to obtain the market correction value V_market.

[0027] ,

[0028] Wherein, S_market is the market supply and demand balance factor, and D_tech is the technology iteration depreciation factor.

[0029] In one alternative approach, the pricing step includes calculating the price P of the photovoltaic power plant using a formula:

[0030] ,

[0031] Where V_market is the market correction value, PV(C_future) is the future maintenance cost, C_future is the future failure probability, and μ is the seller's profit margin.

[0032] In an alternative approach, the method further includes:

[0033] A deep reinforcement learning algorithm is used to construct a sales timing decision model for the photovoltaic power station. Based on the pricing of the photovoltaic power station and the sales timing decision model, the optimal sales time for the photovoltaic power station is determined. The sales timing decision model determines the optimal sales time t by maximizing the cumulative discount reward. Where γ is the discount factor and R(t) is the reward function, which is obtained based on the pricing, cumulative holding cost, opportunity cost and penalty term of the photovoltaic power station.

[0034] According to another aspect of the present invention, a dynamic pricing device for distributed photovoltaic power plants is provided, the device comprising:

[0035] The comprehensive health index determination module is used to collect multi-source heterogeneous data from distributed photovoltaic power plants and determine the comprehensive health index of the photovoltaic power plants based on the multi-source heterogeneous data. The multi-source heterogeneous data includes at least the equipment data, operation data, environmental data, historical maintenance data, performance evaluation data, and market data of the photovoltaic power plants.

[0036] The remaining useful life prediction module is used to obtain the historical health status sequence of the photovoltaic power station equipment, and predict the remaining useful life of the equipment based on the health status sequence and a preset sequence prediction model.

[0037] A construction module is used to build a dynamic pricing model based on the comprehensive health index and the remaining lifespan. Specifically, the construction module is used for:

[0038] The basic value of the photovoltaic power station is determined based on the remaining useful life.

[0039] The basic value is adjusted based on the comprehensive health index to obtain the health-adjusted value;

[0040] Obtain the market supply and demand balance factor and the technology iteration depreciation factor of the photovoltaic power station, and perform market adjustment based on the market supply and demand balance factor, the technology iteration depreciation factor and the health adjustment value to obtain the market adjustment value;

[0041] The pricing module is used to predict the future failure probability of the photovoltaic power station, determine the future maintenance cost of the photovoltaic power station based on the future failure probability, and determine the price of the photovoltaic power station based on the future maintenance cost and the market correction value in the pricing model.

[0042] According to another aspect of the present invention, a computer device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; the memory is used to store at least one executable instruction, wherein the executable instruction causes the processor to perform the method described above.

[0043] According to another aspect of the present invention, a computer-readable storage medium is provided, the storage medium storing at least one executable instruction, which, when executed on a computer device, causes the computer device to perform the method described above.

[0044] The dynamic pricing model of this invention comprehensively considers four dimensions: the lifespan of the photovoltaic power station, its health status, market factors, and future maintenance costs. This results in more accurate pricing, with a very small deviation between the pricing result and the actual market transaction price, which is significantly better than traditional static valuation methods. In addition, this invention uses transfer learning technology, which allows the model to quickly adapt to different types and sizes of distributed photovoltaic power stations. It only requires a small number of samples for fine-tuning to achieve high accuracy, making it widely applicable and highly efficient.

[0045] The above description is merely an overview of the technical solutions of the embodiments of the present invention. In order to better understand the technical means of the embodiments of the present invention and to implement them in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the embodiments of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0046] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0047] Figure 1 A flowchart illustrating the dynamic pricing method for distributed photovoltaic power plants provided in an embodiment of the present invention is shown.

[0048] Figure 2 A schematic diagram is shown illustrating how a temporal convolutional network model predicts the remaining useful life of a photovoltaic power plant.

[0049] Figure 3 A schematic diagram of the structure of the dynamic pricing device for distributed photovoltaic power plants provided in an embodiment of the present invention is shown.

[0050] Figure 4 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention is shown. Detailed Implementation

[0051] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0052] Figure 1 This diagram illustrates a flow chart of the dynamic pricing method for distributed photovoltaic power plants provided in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0053] Step S10 for determining the comprehensive health index involves collecting multi-source heterogeneous data from distributed photovoltaic power stations and determining the comprehensive health index of the photovoltaic power station based on the multi-source heterogeneous data. The multi-source heterogeneous data includes at least the equipment data, operation monitoring data, environmental data, historical maintenance data, performance evaluation data, and market data of the photovoltaic power station.

[0054] The data includes: equipment data such as installed capacity, equipment model, commissioning time, design life, and initial investment cost; operational monitoring data such as photovoltaic module output power, voltage, current, temperature, inverter efficiency, power factor, harmonic content, transformer load rate, and temperature rise; environmental data such as irradiance, ambient temperature, humidity, wind speed, rainfall, and air quality index; historical maintenance data such as fault records, repair time, replaced parts, and maintenance costs; performance evaluation data such as power generation, performance ratio, equivalent utilization hours, and degradation rate; and market data such as transaction prices of similar photovoltaic power plants, market prices of new equipment, electricity pricing policies, and subsidy policies.

[0055] Furthermore, after collecting multi-source heterogeneous data, it is preprocessed, including:

[0056] Data cleaning: Identify and process outliers, missing values, and noisy data, and use the Isolation Forest algorithm to detect outliers;

[0057] Data imputation: Using time-series-based imputation methods to preserve the temporal continuity of data for missing data;

[0058] Data standardization: Perform Z-score standardization or Min-Max normalization on features of different dimensions;

[0059] Feature engineering: Constructing derived features, such as component power degradation rate, inverter efficiency degradation rate, cumulative runtime, start-stop count, load factor, etc.

[0060] In one embodiment, different levels of health indicators can be configured for each item of multi-source heterogeneous data in advance and stored in association. Then, the collected multi-source heterogeneous data is matched with the stored health indicators, and finally, the various health indicators are added together to obtain the comprehensive health index of the photovoltaic power station.

[0061] In another embodiment, the comprehensive health index determination step includes:

[0062] Collect multi-source heterogeneous data from distributed photovoltaic power plants, and determine the component health index, inverter health index, system health index, and reliability index of the photovoltaic power plants based on the multi-source heterogeneous data and a preset deep learning model. The deep learning model is trained using a multi-task learning strategy.

[0063] The weight combination is automatically learned based on the training of the deep learning model. The comprehensive health index H is then determined based on the weight combination, the component health index, the inverter health index, the system health index, and the reliability index.

[0064] ,

[0065] H1 is the component health index, H2 is the inverter health index, H3 is the system health index, and H4 is the reliability index. The weights in the weight combination include w1, w2, w3, and w4, and w1+w2+w3+w4=1.

[0066] This embodiment constructs a deep neural network model to perform multi-dimensional quantitative assessment of the health status of photovoltaic power plants. The deep neural network model can be a DNN, which includes:

[0067] Input layer: Receives multi-dimensional feature vector X after preprocessing of multi-source heterogeneous data;

[0068] Hidden layers: A multi-layer fully connected network structure is adopted, with L layers (L≥3). The number of neurons in each layer decreases sequentially. For example, in a 4-layer fully connected network, the number of neurons in the first hidden layer is 256, the number of neurons in the second hidden layer is 128, the number of neurons in the third hidden layer is 64, and the number of neurons in the fourth hidden layer is 32. The activation function can be the modified linear unit ReLU or its variant Leaky ReLU.

[0069] Attention mechanism layer: Introducing a self-attention mechanism to learn the importance weights of different features to health status, automatically identifying key depletion factors, and improving the interpretability of the assessment;

[0070] Output layer: Outputs scores for multiple health dimensions, including:

[0071] The module health index H1, where H1∈[0,1], represents the performance status of the photovoltaic module;

[0072] Inverter health index H2, H2∈[0,1], H2 represents the operating status of the inverter;

[0073] The system health index H3, where H3∈[0,1], represents the overall health level of the system.

[0074] Reliability index H4, H4∈[0,1]: characterizes the reliability and stability of the system.

[0075] The weights w1, w2, w3, and w4 are automatically learned during training. The deep learning model employs a multi-task learning strategy, and the loss function L_total is:

[0076] ,

[0077] Where L_mse is the mean squared error loss, L_rank is the ranking loss (to ensure that the health score is consistent with the order of the actual state), L_reg is the regularization term, and λ1-λ3 are the corresponding weights.

[0078] This embodiment comprehensively considers more than 20 key factors across multiple dimensions, such as photovoltaic module performance degradation, inverter operating status, and support structure integrity, to obtain a comprehensive health index H, thereby accurately quantifying the multi-dimensional health status of the photovoltaic power station.

[0079] The remaining useful life prediction step S20 involves obtaining the historical health status sequence of the photovoltaic power station equipment, and predicting the remaining useful life of the equipment based on the health status sequence and a preset sequence prediction model.

[0080] The health status sequence can be a sequence with T time steps and d dimensions. The features of the d dimensions can include the health index, health degradation rate, cumulative operating time, and environmental load factor of the photovoltaic power station equipment, etc.

[0081] Furthermore, the sequence prediction model is a temporal convolutional network model (TCN). The TCN model includes multiple concatenated blocks, each containing residual connections and multiple causal dilated convolutional layers with different dilation rates. The multiple causal dilated convolutional layers simultaneously capture the temporal features of the health state sequence across multiple time scales from short-term to long-term. The outputs of the multiple causal dilated convolutional layers are fused together as the output features of the block.

[0082] See also Figure 2 The number of blocks is 3, and each block contains multiple causal dilated convolutional layers with different dilation rates. Figure 2In the block diagram, blocks 1, 2, and 3 can each include causal dilation convolution 1 and causal dilation convolution 2 with different dilation rates. For example, block 1 can include causal dilation convolution 1 and causal dilation convolution 2 with a dilation rate of 1, 3 kernels, and 64 output channels; block 2 can include causal dilation convolution 1 and causal dilation convolution 2 with a dilation rate of 2, 3 kernels, and 128 output channels; and block 3 can include causal dilation convolution 1 and causal dilation convolution 2 with a dilation rate of 4, 3 kernels, and 256 output channels. Using causal dilation convolution ensures that predictions rely only on historical data, and dilation convolution can expand the receptive field to capture long-term dependencies. Furthermore, each block contains residual connections to mitigate the gradient vanishing problem. In the feature fusion layer, temporal global features can be extracted using global average pooling. In the output layer, the output includes:

[0083] Remaining useful life R: The predicted time from the current moment to the moment of failure of the equipment, in years or hours;

[0084] Future health decline curve f(t): predicts the trend of health index changes over a future period of time;

[0085] Prediction confidence interval: The Monte Carlo technique is used to estimate the uncertainty of the prediction and output the 95% confidence interval of the remaining useful life.

[0086] This embodiment uses a temporal convolutional network model with the above structure to scientifically and accurately predict the remaining service life of photovoltaic power plants.

[0087] Furthermore, the temporal convolutional network model is trained using a loss function that incorporates physical laws, wherein the loss function L is: Where L_pred is the prediction error, L_physics is the physical constraint term, and γ is the weight parameter. The physical constraint term is constructed based on the physical degradation model of the device, and the performance degradation of the device follows a Weibull distribution.

[0088] To further improve prediction accuracy, the temporal convolutional network model is trained by incorporating a loss function based on physical laws. Assuming that the performance degradation of photovoltaic modules or equipment follows a Weibull distribution, the probability density function of the Weibull distribution is embedded as prior knowledge into the loss function L for training.

[0089] ,

[0090] L_physics is a physical constraint term that ensures the prediction results conform to the physical degradation law of the equipment.

[0091] Step S30, constructing a pricing model, involves building a dynamic pricing model based on the comprehensive health index and the remaining lifespan. This step includes:

[0092] The basic value of the photovoltaic power station is determined based on the remaining useful life.

[0093] The basic value is adjusted based on the comprehensive health index to obtain the health-adjusted value;

[0094] The market dynamic factors of the photovoltaic power station are obtained, and the market is adjusted based on the market dynamic factors and the health adjustment value to obtain the market adjustment value.

[0095] The fundamental value of a photovoltaic power station is measured primarily by its remaining useful life. The longer the remaining useful life, the greater the fundamental value, and vice versa.

[0096] In one embodiment, determining the basic value of the photovoltaic power station based on the remaining useful life includes: obtaining a lifespan value function based on the remaining useful life, the design life of the photovoltaic power station, and a deep learning model; and determining the basic value based on the lifespan value function and the market price of a newly built photovoltaic power station of the same scale.

[0097] The fundamental value V is calculated using the following formula: Where V_new is the market price of a newly built photovoltaic power station of the same scale, f_life is the life value function learned through a deep learning model, R is the remaining lifespan, and L_design is the design lifespan of the photovoltaic power station; , where a is the fitting parameter, a∈[0.6,0.8], obtained by fitting historical transaction data, and reflects the nonlinear characteristics of the life loss of photovoltaic power plants.

[0098] This embodiment can accurately measure the fundamental value of a photovoltaic power station by using a lifetime value function based on the remaining useful life and the market price of a newly built photovoltaic power station of the same scale.

[0099] In this embodiment, in addition to the basic value of the photovoltaic power station, the health status of the photovoltaic power station is also considered. The basic value is corrected according to the comprehensive health index. If the comprehensive health index is large and the stability is good, the corrected health value is larger. If the comprehensive health index is small and the stability is poor, the corrected health value is smaller.

[0100] In one embodiment, adjusting the baseline value based on the comprehensive health index to obtain a health-adjusted value includes: determining a health adjustment function based on the comprehensive health index and its variance; determining the health-adjusted value based on the health adjustment function and the baseline value; and specifying the health-adjusted value as the adjusted value. ,

[0101] Wherein, V is the basic value, g_health is the health correction function, H is the comprehensive health index, and σ is the variance of the comprehensive health index. The variance reflects the stability of the health status of the photovoltaic power station, and σ can be obtained from the output of the deep learning model in the above comprehensive health index determination steps.

[0102] Where β is the health sensitivity coefficient, H_ref is the reference health level (generally H_ref=0.7), and k is the fluctuation penalty coefficient, which can be obtained through deep learning.

[0103] This embodiment, based on the health correction value of a photovoltaic power station, also considers market dynamic factors of the photovoltaic power station. The health correction value is further corrected according to the market dynamic factors, including market supply and demand balance factors and technology iteration depreciation factors. If the market supply and demand are more balanced and the technology iteration depreciation rate is slower, the corrected market correction value will be greater, and vice versa.

[0104] In one embodiment, the market supply and demand balance factor is predicted based on a preset time-series network model. The step of obtaining the market dynamic factor of the photovoltaic power station and performing market correction based on the market dynamic factor and the health correction value to obtain the market correction value includes: performing market correction based on the market supply and demand balance factor, the technology iteration depreciation factor, and the health correction value to obtain the market correction value V_market.

[0105] ,

[0106] Where S_market is the market supply and demand balance factor, and D_tech is the technology iteration depreciation factor. The time series network model can be a Long Short-Term Memory (LSTM) network, S_market = LSTM(j, c, z), where j is the historical trading volume, c is the new production capacity, and z is the policy change; , λ represents the generational gap between photovoltaic power plant technology and current mainstream technology, expressed in years, and λ is the technology iteration rate constant.

[0107] This embodiment further refines the basic value of a photovoltaic power station by incorporating a comprehensive health index and market dynamics factors, thereby obtaining an accurate value for the photovoltaic power station.

[0108] In pricing step S40, the future failure probability of the photovoltaic power station is predicted, the future maintenance cost of the photovoltaic power station is determined based on the future failure probability, and the price of the photovoltaic power station is determined based on the future maintenance cost and the market correction value in the pricing model.

[0109] This embodiment uses a temporal convolutional network model to predict the future failure probability of the photovoltaic power station. P i (t) represents the probability of the i-th type of failure of the photovoltaic equipment or module occurring at time t, C i Let be the maintenance cost for the i-th type of fault.

[0110] Future maintenance costs ,

[0111] Where R is the remaining useful life and r is the discount rate.

[0112] Finally, the pricing P of the photovoltaic power station was calculated using the formula:

[0113] ,

[0114] Where V_market is the market correction value, PV(C_future) is the future maintenance cost, C_future is the future failure probability, and μ is the seller's profit margin. Generally, μ is 5%-15%, determined according to the degree of market competition and the bargaining power of the buyer and seller.

[0115] It is understood that the deep learning model in this embodiment can be a DNN, or other deep learning architectures such as Transformer, ResNet, etc. The sequence prediction model can be a Temporal Convolutional Network (TCN), or LSTM, GRU, or Transformer, etc., and is not limited here. This embodiment of the invention can be widely applied to various scenarios such as photovoltaic power plant asset transactions, merger and acquisition valuation, asset securitization, and insurance pricing.

[0116] The dynamic pricing model of this invention comprehensively considers four dimensions: the lifespan of the photovoltaic power station, its health status, market factors, and future maintenance costs. This results in more accurate pricing, with a very small deviation from the actual market transaction price (less than 8%), which is significantly better than traditional static valuation methods. Furthermore, this invention employs transfer learning technology, allowing the model to quickly adapt to different types and scales of distributed photovoltaic power stations. It requires only a small number of samples for fine-tuning to achieve high accuracy, making it widely applicable and highly efficient.

[0117] In an optional embodiment, the dynamic pricing method for distributed photovoltaic power plants further includes:

[0118] A deep reinforcement learning algorithm is used to construct a sales timing decision model for the photovoltaic power station. Based on the pricing of the photovoltaic power station and the sales timing decision model, the optimal sales time for the photovoltaic power station is determined. The sales timing decision model determines the optimal sales time t by maximizing the cumulative discount reward.

[0119] Where γ is the discount factor and R(t) is the reward function, which is obtained based on the pricing, cumulative holding cost, opportunity cost and penalty term of the photovoltaic power station.

[0120] The deep reinforcement learning algorithm in this embodiment can be a deep Q-network (DQN), or it can be replaced with other reinforcement learning algorithms, such as A3C, PPO, etc., which are not limited here.

[0121] Taking deep Q-networks as an example, a decision-making model for the timing of photovoltaic power plant sales is constructed. In this decision-making model, the state space, action space and reward function are defined.

[0122] 1. Definition of state space:

[0123] S(t)=[H(t), R(t), P_market(t), C_hold(t), dH / dt],

[0124] Where H(t) is the current health index, R(t) is the current remaining useful life, P_market(t) is the current market price trend, C_hold(t) is the cumulative holding cost, and dH / dt is the health decay rate.

[0125] 2. Definition of Action Space:

[0126] A = {a0: sell immediately, a1: hold for 1 month, a2: hold for 3 months, a3: hold for 6 months}. Of course, the action space can also be defined as other actions, which are not limited here.

[0127] 3. Design of reward function R(t):

[0128] R(t)=P(t)-C_hold(t)-C_opportunity(t)-Penalty(t),

[0129] Where P(t) is the current pricing of the photovoltaic power station, C_opportunity(t) is the current opportunity cost (i.e., the time value of money), and Penalty(t) is the penalty for health being below the threshold.

[0130] 4. Decision-making algorithm based on the selling timing decision-making model:

[0131] The optimal selling time t is determined by maximizing the cumulative discount reward, with the following constraints:

[0132] Health constraint: H(t) ≥ H_min, generally, H_min = 0.5;

[0133] Remaining useful life constraint: R(t)≥R_min, generally, R_min=5 years;

[0134] Market liquidity constraint: There are enough potential buyers (e.g., more than 5 buyers) that the transaction can be completed within a reasonable time.

[0135] This embodiment is based on a photovoltaic power plant sales timing decision model using deep Q-networks. It can scientifically and reasonably provide optimal sales timing suggestions for photovoltaic power plants under multiple constraints.

[0136] For example, the valuation of a rooftop distributed photovoltaic power station in an industrial park will be used as an example for illustration.

[0137] 1. Basic information about a rooftop distributed photovoltaic power station in an industrial park is as follows:

[0138] Project parameters Specific values Installed capacity 2MW Commissioning time June 2018 Runtime 6.5 years Design life 25 years Component type Monocrystalline silicon PERC module, 345W Inverter type String inverter, 50kW Initial investment 11 million yuan (including subsidies)

[0139] 2. Data Acquisition and Preprocessing:

[0140] The following data were collected through the photovoltaic power plant's SCADA system and a third-party monitoring platform:

[0141] 6.5 years of historical operating data, including power generation, voltage, current, and component temperature every 5 minutes, totaling approximately 6.8 million records;

[0142] Meteorological data: Data such as irradiance, ambient temperature, and wind speed provided by the local weather station;

[0143] Maintenance records: Detailed records of 8 scheduled inspections, 3 fault repairs, and 2 inverter replacements;

[0144] On-site testing data: results from infrared thermal imaging, EL testing, IV curve testing, etc.

[0145] Data preprocessing steps:

[0146] 0.3% of outlier data points (such as non-zero power at night or values ​​exceeding physical limits) were identified and removed.

[0147] For the missing data (accounting for 1.2%), a combination of linear interpolation and LSTM interpolation was used;

[0148] Thirty-two derived features were calculated, including performance ratio (PR), decay rate, temperature correction factor, and load factor.

[0149] 3. Health status assessment:

[0150] Input layer: 32 feature dimensions;

[0151] Hidden layers: 4 fully connected layers with 256-128-64-32 neurons and LeakyReLU activation function;

[0152] Attention mechanism layer: learns feature weights;

[0153] Output layer: 4 health dimensions score.

[0154] Evaluation results:

[0155] Health Dimension score Weight Key influencing factors Component Health Index H1 0.82 0.40 Power attenuation, hot spots Inverter Health Index H2 0.75 0.30 Decreased efficiency, number of start-stop cycles System Health Index H3 0.88 0.20 Line loss, obstruction Reliability Index H4 0.85 0.10 Failure rate, MTBF Comprehensive Health Index H 0.82 - -

[0156] 4. Remaining useful life prediction:

[0157] Construct a TCN model, input a 78-month historical health status sequence, and predict the following results:

[0158] The remaining useful life R = 17.3 years;

[0159] 95% confidence interval: [16.5 years, 18.2 years];

[0160] The projected future decline curve shows that the health index will decline at a rate of approximately 0.8% per year.

[0161] 5. Constructing a pricing model and setting prices:

[0162] Basic value calculation:

[0163] V_new = 8.5 million yuan (current market price of a 2MW newly built power plant);

[0164] a=0.72 (fitted using historical data);

[0165] f_life(17.3, 25)=0.72×(17.3 / 25)+0.28×(17.3 / 25)²=0.498+0.134=0.632,

[0166] The basic value V = 8.5 million × 0.632 = 5.372 million yuan.

[0167] Health status correction:

[0168] β=0.25, k=0.15, σ=0.03;

[0169] g_health(0.82,0.03)=1+0.25×(0.82-0.7)-0.15×0.03=1+0.03-0.0045=1.025;

[0170] The health-corrected value V_health = 5.372 million × 1.025 = 5.506 million yuan.

[0171] Market dynamic factors:

[0172] Market supply and demand balance factor S_market = 1.08 (LSTM predicts strong current market demand).

[0173] Δt = 3 years (technological gap), λ = 0.05;

[0174] The technology iteration depreciation factor D_tech = exp(-0.05 × 3) = 0.861;

[0175] The market correction value V_market = 5,506,000 × 1.08 × 0.861 = 5,118,000 yuan.

[0176] Future maintenance costs:

[0177] TCN forecasts maintenance costs for the next 17.3 years (discount rate r=6%):

[0178] PV(C_future) = 325,000 yuan;

[0179] Final pricing: μ = 10% (profit margin), P = (5.118 million - 325,000) × (1 + 0.1) = 5.272 million yuan.

[0180] Compared with existing technologies, the pricing of the rooftop distributed photovoltaic power station in this industrial park using the straight-line depreciation method, the unit capacity depreciation method, and the discounted cash flow method is 6.28 million yuan, 4.8 million yuan, and 5.58 million yuan, respectively. However, the actual transaction price of the rooftop distributed photovoltaic power station in this industrial park is 5.25 million yuan, with a deviation of 0.4%, which is the smallest deviation.

[0181] Figure 3 A schematic diagram of the structure of a dynamic pricing device for a distributed photovoltaic power station according to an embodiment of the present invention is shown. Figure 3 As shown, the device includes:

[0182] The comprehensive health index determination module 301 is used to collect multi-source heterogeneous data of distributed photovoltaic power stations and determine the comprehensive health index of the photovoltaic power station based on the multi-source heterogeneous data. The multi-source heterogeneous data includes at least the equipment data, operation data, environmental data, historical maintenance data, performance evaluation data and market data of the photovoltaic power station.

[0183] The remaining useful life prediction module 302 is used to obtain the historical health status sequence of the photovoltaic power station equipment, and predict the remaining useful life of the equipment based on the health status sequence and a preset sequence prediction model.

[0184] Construction module 303 is used to construct a dynamic pricing model based on the comprehensive health index and the remaining lifespan. Specifically, construction module 303 is used for:

[0185] The basic value of the photovoltaic power station is determined based on the remaining useful life.

[0186] The basic value is adjusted based on the comprehensive health index to obtain the health-adjusted value;

[0187] Obtain the market supply and demand balance factor and the technology iteration depreciation factor of the photovoltaic power station, and perform market adjustment based on the market supply and demand balance factor, the technology iteration depreciation factor and the health adjustment value to obtain the market adjustment value;

[0188] The pricing module 304 is used to predict the future failure probability of the photovoltaic power station, determine the future maintenance cost of the photovoltaic power station based on the future failure probability, and determine the price of the photovoltaic power station based on the future maintenance cost and the market correction value in the pricing model.

[0189] The embodiments of the dynamic pricing device for distributed photovoltaic power plants are basically the same as those of the dynamic pricing method for distributed photovoltaic power plants described above, and can be referred to the above embodiments.

[0190] Figure 4 The diagram shows a structural schematic of an embodiment of the computer device of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computer device.

[0191] like Figure 4 As shown, the computer device may include: a processor 402, a communications interface 404, a memory 406, and a communications bus 408.

[0192] The processor 402, communication interface 404, and memory 406 communicate with each other via communication bus 408. Communication interface 404 is used to communicate with other computer devices, such as clients or other server network elements. The processor 402 executes program 410, specifically performing the relevant steps described above in the computer device embodiment.

[0193] Specifically, program 410 may include program code, which includes computer-executable instructions.

[0194] Processor 402 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0195] Memory 406 is used to store program 410. Memory 406 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0196] Specifically, program 410 can be called by processor 402 to cause the computer device to perform the following operations:

[0197] The comprehensive health index determination steps involve collecting multi-source heterogeneous data from distributed photovoltaic power plants, and determining the comprehensive health index of the photovoltaic power plants based on the multi-source heterogeneous data. The multi-source heterogeneous data includes at least the equipment data, operation monitoring data, environmental data, historical maintenance data, performance evaluation data, and market data of the photovoltaic power plants.

[0198] The remaining useful life prediction step involves obtaining the historical health status sequence of the photovoltaic power station equipment, and predicting the remaining useful life of the equipment based on the health status sequence and a preset sequence prediction model.

[0199] The step of constructing a pricing model involves building a dynamic pricing model based on the comprehensive health index and the remaining lifespan. This step includes:

[0200] The basic value of the photovoltaic power station is determined based on the remaining useful life.

[0201] The basic value is adjusted based on the comprehensive health index to obtain the health-adjusted value;

[0202] The market dynamic factors of the photovoltaic power station are obtained, and the market is adjusted based on the market dynamic factors and the health adjustment value to obtain the market adjustment value.

[0203] The pricing process involves predicting the future failure probability of the photovoltaic power station, determining the future maintenance cost of the photovoltaic power station based on the future failure probability, and determining the price of the photovoltaic power station based on the future maintenance cost and the market correction value in the pricing model.

[0204] In one alternative approach, the comprehensive health index determination step includes:

[0205] Collect multi-source heterogeneous data from distributed photovoltaic power plants, and determine the component health index, inverter health index, system health index, and reliability index of the photovoltaic power plants based on the multi-source heterogeneous data and a preset deep learning model. The deep learning model is trained using a multi-task learning strategy.

[0206] The weight combination is automatically learned based on the training of the deep learning model. The comprehensive health index H is then determined based on the weight combination, the component health index, the inverter health index, the system health index, and the reliability index.

[0207] ,

[0208] H1 is the component health index, H2 is the inverter health index, H3 is the system health index, and H4 is the reliability index. The weights in the weight combination include w1, w2, w3, and w4, and w1+w2+w3+w4=1.

[0209] In one alternative approach, the sequence prediction model is a temporal convolutional network model, which includes multiple concatenated blocks. Each block contains residual connections and multiple causal dilated convolutional layers with different dilation rates. The multiple causal dilated convolutional layers simultaneously capture the temporal features of the health state sequence at multiple time scales from short-term to long-term. The outputs of the multiple causal dilated convolutional layers are fused together as the output features of the block.

[0210] In one alternative approach, the temporal convolutional network model is trained using a loss function that incorporates physical laws, wherein the loss function L is: Where L_pred is the prediction error, L_physics is the physical constraint term, and γ is the weight parameter. The physical constraint term is constructed based on the physical degradation model of the device, and the performance degradation of the device follows a Weibull distribution.

[0211] In one alternative approach, determining the fundamental value of the photovoltaic power station based on the remaining useful life includes:

[0212] Based on the remaining useful life, the design life of the photovoltaic power station and the life value function learned by the deep learning model, the basic value is determined based on the life value function and the market price of a newly built photovoltaic power station of the same scale.

[0213] Among them, the basic value V_new is the market price of the newly built photovoltaic power station of the same scale, f_life is the lifetime value function, R is the remaining lifetime, and L_design is the design lifetime of the photovoltaic power station.

[0214] , where a is the fitting parameter;

[0215] The step of adjusting the baseline value based on the comprehensive health index to obtain a health-adjusted value includes: determining a health adjustment function based on the comprehensive health index and its variance; determining the health-adjusted value based on the health adjustment function and the baseline value; and specifying the health-adjusted value as the adjusted value. ,

[0216] Where g_health is the health correction function, H is the comprehensive health index, and σ is the variance of the comprehensive health index;

[0217] The market dynamic factors include market supply and demand balance factors and technology iteration depreciation factors. The market supply and demand balance factors are predicted based on a preset time-series network model. The process of obtaining the market dynamic factors of the photovoltaic power station and performing market corrections based on the market dynamic factors and the health correction value to obtain the market correction value includes: performing market corrections based on the market supply and demand balance factors, the technology iteration depreciation factor, and the health correction value to obtain the market correction value V_market. ,

[0218] Wherein, S_market is the market supply and demand balance factor, and D_tech is the technology iteration depreciation factor.

[0219] In one alternative approach, the pricing step includes calculating the price P of the photovoltaic power plant using a formula:

[0220] P=[V_market-PV(C_future)]×(1+μ),

[0221] Where V_market is the market correction value, PV(C_future) is the future maintenance cost, C_future is the future failure probability, and μ is the seller's profit margin.

[0222] In an alternative approach, the method further includes:

[0223] A deep reinforcement learning algorithm is used to construct a sales timing decision model for the photovoltaic power station. Based on the pricing of the photovoltaic power station and the sales timing decision model, the optimal sales time for the photovoltaic power station is determined. The sales timing decision model determines the optimal sales time t by maximizing the cumulative discount reward.

[0224] Where γ is the discount factor and R(t) is the reward function, which is obtained based on the pricing, cumulative holding cost, opportunity cost and penalty term of the photovoltaic power station.

[0225] This invention provides a computer-readable storage medium storing at least one executable instruction that, when executed on a computer device, causes the computer device to perform any of the above-described method embodiments.

[0226] This invention provides a computer program that can be invoked by a processor to cause a computer device to execute any of the above-described method embodiments.

[0227] This invention provides a computer program product, which includes a computer program stored on a computer-readable storage medium. The computer program includes program instructions that, when executed on a computer, cause the computer to perform any of the above-described method embodiments.

[0228] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, the embodiments of the present invention are not directed to any particular programming language. It should be understood that the content of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0229] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0230] Similarly, it should be understood that, in order to streamline the invention and aid in understanding one or more of the various aspects of the invention, features of the embodiments of the invention are sometimes grouped together in a single embodiment, figure, or description thereof in the above description of exemplary embodiments of the invention. However, this disclosure should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim.

[0231] Those skilled in the art will understand that modules in the computer device of the embodiments can be adaptively modified and placed in one or more computer devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or computer device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0232] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the order of execution.

Claims

1. A dynamic pricing method for distributed photovoltaic power plants, characterized in that, The method includes: The comprehensive health index determination steps involve collecting multi-source heterogeneous data from distributed photovoltaic power plants, and determining the comprehensive health index of the photovoltaic power plants based on the multi-source heterogeneous data. The multi-source heterogeneous data includes at least the equipment data, operation monitoring data, environmental data, historical maintenance data, performance evaluation data, and market data of the photovoltaic power plants. The remaining useful life prediction step involves obtaining the historical health status sequence of the photovoltaic power station equipment, and predicting the remaining useful life of the equipment based on the health status sequence and a preset sequence prediction model. The step of constructing a pricing model involves building a dynamic pricing model based on the comprehensive health index and the remaining lifespan. This step includes: The basic value of the photovoltaic power station is determined based on the remaining useful life. The basic value is adjusted based on the comprehensive health index to obtain the health-adjusted value; The market dynamic factors of the photovoltaic power station are obtained, and the market is adjusted based on the market dynamic factors and the health adjustment value to obtain the market adjustment value. The pricing process involves predicting the future failure probability of the photovoltaic power station, determining the future maintenance cost of the photovoltaic power station based on the future failure probability, and determining the price of the photovoltaic power station based on the future maintenance cost and the market correction value in the pricing model. The determination of the basic value of the photovoltaic power station based on the remaining useful life includes: Based on the remaining useful life, the design life of the photovoltaic power station and the life value function learned by the deep learning model, the basic value is determined based on the life value function and the market price of a newly built photovoltaic power station of the same scale. Wherein, the basic value V = V_new * f_life(R, L_design), V_new is the market price of the newly built photovoltaic power station of the same scale, f_life is the life value function, R is the remaining lifespan, and L_design is the design lifespan of the photovoltaic power station; f_life(R,L_design)=a*(R / L_design)+(1-a)*(R / L_design)^2, where a is the fitting parameter; The step of adjusting the baseline value based on the comprehensive health index to obtain the health-adjusted value includes: determining a health adjustment function based on the comprehensive health index and its variance; determining the health-adjusted value based on the health adjustment function and the baseline value; and defining the health-adjusted value as V_health = V*g_health(H,σ). Where g_health is the health correction function, H is the comprehensive health index, and σ is the variance of the comprehensive health index; The market dynamic factors include market supply and demand balance factors and technology iteration depreciation factors. The market supply and demand balance factors are predicted based on a preset time-series network model. The process of obtaining the market dynamic factors of the photovoltaic power station and performing market corrections based on the market dynamic factors and the health correction value to obtain the market correction value includes: performing market corrections based on the market supply and demand balance factors, the technology iteration depreciation factor, and the health correction value to obtain the market correction value V_market. V_market=V_health*S_market*D_tech, Wherein, S_market is the market supply and demand balance factor, and D_tech is the technology iteration depreciation factor.

2. The method according to claim 1, characterized in that, The steps for determining the comprehensive health index include: Collect multi-source heterogeneous data from distributed photovoltaic power plants, and determine the component health index, inverter health index, system health index, and reliability index of the photovoltaic power plants based on the multi-source heterogeneous data and a preset deep learning model. The deep learning model is trained using a multi-task learning strategy. The weight combination is automatically learned based on the training of the deep learning model. The comprehensive health index H is then determined based on the weight combination, the component health index, the inverter health index, the system health index, and the reliability index. ， H1 is the component health index, H2 is the inverter health index, H3 is the system health index, and H4 is the reliability index. The weights in the weight combination include w1, w2, w3, and w4, and w1+w2+w3+w4=1.

3. The method according to claim 1, characterized in that, The sequence prediction model is a temporal convolutional network model, which includes multiple concatenated blocks. Each block contains residual connections and multiple causal dilated convolutional layers with different dilation rates. The multiple causal dilated convolutional layers simultaneously capture the temporal features of the health state sequence at multiple time scales from short-term to long-term. The outputs of the multiple causal dilated convolutional layers are fused together as the output features of the block.

4. The method according to claim 3, characterized in that, The temporal convolutional network model is trained using a loss function that incorporates physical laws. The loss function L is: L = L_pred + γ * L_physics, where L_pred is the prediction error, L_physics is the physical constraint term, and γ is the weight parameter. The physical constraint term is constructed based on the physical degradation model of the device, and the performance degradation of the device follows a Weibull distribution.

5. The method according to claim 1, characterized in that, The pricing step includes calculating the price P of the photovoltaic power station using a formula: ， Where V_market is the market correction value, PV(C_future) is the future maintenance cost, C_future is the future failure probability, and μ is the seller's profit margin.

6. The method according to claim 1, characterized in that, The method further includes: A deep reinforcement learning algorithm is used to construct a sales timing decision model for the photovoltaic power station. Based on the pricing of the photovoltaic power station and the sales timing decision model, the optimal sales time for the photovoltaic power station is determined. The sales timing decision model determines the optimal sales time t by maximizing the cumulative discount reward. Where γ is the discount factor and R(t) is the reward function, which is obtained based on the pricing, cumulative holding cost, opportunity cost and penalty term of the photovoltaic power station.

7. A dynamic pricing device for distributed photovoltaic power plants, characterized in that, The device includes: The comprehensive health index determination module is used to collect multi-source heterogeneous data from distributed photovoltaic power plants and determine the comprehensive health index of the photovoltaic power plants based on the multi-source heterogeneous data. The multi-source heterogeneous data includes at least the equipment data, operation data, environmental data, historical maintenance data, performance evaluation data, and market data of the photovoltaic power plants. The remaining useful life prediction module is used to obtain the historical health status sequence of the photovoltaic power station equipment, and predict the remaining useful life of the equipment based on the health status sequence and a preset sequence prediction model. A construction module is used to build a dynamic pricing model based on the comprehensive health index and the remaining lifespan. Specifically, the construction module is used for: The basic value of the photovoltaic power station is determined based on its remaining useful life. Based on the remaining useful life, the design life of the photovoltaic power station and the life value function learned by the deep learning model, the basic value is determined based on the life value function and the market price of a newly built photovoltaic power station of the same scale. Wherein, the basic value V = V_new * f_life(R, L_design), V_new is the market price of the newly built photovoltaic power station of the same scale, f_life is the life value function, R is the remaining lifespan, and L_design is the design lifespan of the photovoltaic power station; f_life(R,L_design)=a*(R / L_design)+(1-a)*(R / L_design)^2, where a is the fitting parameter; The baseline value is adjusted based on the comprehensive health index to obtain the health-adjusted value: a health adjustment function is determined based on the comprehensive health index and its variance; the health-adjusted value is then determined based on the health adjustment function and the baseline value; the health-adjusted value V_health = V*g_health(H,σ). Where g_health is the health correction function, H is the comprehensive health index, and σ is the variance of the comprehensive health index; Obtain the market supply and demand balance factor and technology iteration depreciation factor of the photovoltaic power station. Based on the market supply and demand balance factor, the technology iteration depreciation factor, and the health correction value, perform market correction to obtain the market correction value V_market. The market supply and demand balance factor is predicted based on a preset time-series network model. V_market=V_health*S_market*D_tech, Wherein, S_market is the market supply and demand balance factor, and D_tech is the technology iteration depreciation factor; The pricing module is used to predict the future failure probability of the photovoltaic power station, determine the future maintenance cost of the photovoltaic power station based on the future failure probability, and determine the price of the photovoltaic power station based on the future maintenance cost and the market correction value in the pricing model.

8. A computer device, characterized in that, include: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction that causes the processor to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The storage medium stores at least one executable instruction, which, when executed on a computer device, causes the computer device to perform the method as described in any one of claims 1-6.

Images