Charging station loss prediction method based on LSTM deep learning method

Abstract

The invention discloses a charging station loss prediction method based on an LSTM deep learning method, and relates to the technical field of charging stations, and the method comprises the following steps: 1, collecting data including real-time power load data, a power grid topological structure and time sequence data; 2, feature engineering: carrying out preprocessing and feature extraction on the data, and extracting expert knowledge features; step 3, establishing an LSTM model in combination with a PPO algorithm; 4, performing hyper-parameter tuning on the LSTM model through a random search method; and step 5, the optimized miniaturized model is then deployed to each charging station, and local real-time energy loss prediction is realized. In order to further improve the prediction accuracy, expert knowledge is creatively integrated in a feature engineering part, multi-dimensional loss feature recognition is carried out, and the performance of the model in processing a complex nonlinear relation is improved by combining the input of an electric power system expert, recognizing factors such as equipment aging and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of energy conservation and environmental protection, and in particular to a charging station loss prediction method based on an LSTM deep learning method. Background Art

[0002] With the rapid increase in the number of electric vehicles, the demand for electric vehicle charging infrastructure is also growing. As the main facility for electric vehicle energy replenishment, the construction and operation of electric vehicle charging stations face many challenges, among which the energy loss problem is particularly prominent. During the charging process of electric vehicles, energy loss mainly comes from several aspects:

[0003] 1. Power transmission loss: When power is transmitted from the power grid to the charging station, it will generate losses due to the resistance and inductance of cables and power equipment (such as transformers, rectifiers, etc.). 2. Power conversion loss: The power electronic equipment inside the charging station, such as inverters, DC-DC converters, etc., will generate losses when converting power from the power grid into electricity suitable for electric vehicle batteries. 3. Charging process loss: During the charging process of electric vehicles, factors such as the internal resistance of the battery, charging current, and battery management system will cause energy loss. 4. Standby and maintenance loss: When the charging station is not in working state, some equipment still needs to remain in standby state to respond to charging needs at any time, which will also generate certain energy losses. 5. Impact of environmental factors: External conditions such as ambient temperature and humidity will also affect the energy loss of the charging station.

[0004] Accurately predicting the energy loss of charging stations can help charging stations optimize their decisions, reduce operating costs, improve charging efficiency, enhance user experience, and promote the popularization and promotion of electric vehicles. Therefore, how to accurately predict the energy loss of charging stations has become an urgent problem to be solved. However, the existing energy loss prediction of charging stations still has many shortcomings:

[0005] 1. Low prediction accuracy: Traditional energy loss prediction methods mainly rely on simple statistical models or linear regression models, which are difficult to capture complex nonlinear relationships, resulting in low prediction accuracy.

[0006] 2. Insufficient real-time performance: The energy loss of charging stations is affected by many dynamic factors, such as instantaneous load changes and power grid fluctuations. Existing methods are difficult to obtain and process these data in real time, resulting in delayed prediction results and difficulty in meeting the needs of real-time optimization of charging stations. For example, traditional methods require long-term data accumulation and batch processing, which cannot reflect changes in real-time data in a timely manner.

[0007] 3. Limited data processing capabilities: A large amount of high-dimensional data, such as voltage, current, power, frequency, etc., is generated during the operation of charging stations. These data are characterized by high dimensionality, nonlinearity, and complexity. Traditional statistical methods and machine learning methods often show problems of low processing efficiency and high computational complexity when processing these large-scale data, making it difficult to fully utilize the information in the data for accurate predictions.

[0008] Therefore, how to accurately predict the energy loss of charging stations to help operators optimize decisions, reduce operating costs, and improve charging efficiency has become an urgent problem to be solved. In recent years, with the development of deep learning technology, especially the emergence of long short-term memory (LSTM) networks, new ideas have been provided to solve the problem of energy loss prediction of charging stations. The LSTM model can process time series data and capture complex nonlinear relationships, and has high potential in energy loss prediction. Summary of the invention

[0009] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0010] In view of the problems existing in the above-mentioned existing charging station loss prediction method based on LSTM deep learning method, the present invention is proposed.

[0011] In order to solve the above technical problems, the present invention provides the following technical solutions: a charging station loss prediction method based on LSTM deep learning method, the method comprising the following steps:

[0012] Step 1: Data collection, including but not limited to: real-time power load data, power grid topology, time series data, environmental factors, and equipment status;

[0013] Step 2: Feature engineering: preprocessing and feature extraction of data, smoothing fluctuations, processing abnormal data, extracting expert knowledge features, and enhancing the predictive ability and interpretability of the model;

[0014] Step 3: Combine the PPO algorithm to establish an LSTM model. The LSTM model is used to capture the long-term dependencies in the energy loss time series data of the charging station. Combined with the PPO algorithm, the state parameters of the model are optimized to improve the prediction accuracy and convergence speed.

[0015] Step 4: Hyperparameter tuning: Use the random search method to tune the hyperparameters of the LSTM model and find the optimal parameter combination to improve model performance.

[0016] Step 5: Model deployment and application, using a centralized training and decentralized prediction architecture, collects historical data from each charging station and conducts centralized training on a central server; the optimized miniaturized model is then deployed to each charging station to achieve local real-time energy loss prediction.

[0017] As a preferred solution of the charging station loss prediction method based on the LSTM deep learning method described in the present invention, the real-time power load data collection method is to collect real-time power load data such as current, voltage, power factor and frequency through smart meters, remote terminal units and smart transformers;

[0018] The grid topology information acquisition method is: extracting the connection relationship and configuration of each part of the grid from the grid geographic information system and asset management system;

[0019] The time series data selection method is as follows: data is selected from 54 variables such as historical power flow data, load demand, line loss, and charging pile utilization rate, and input in time series to ensure the long-term availability and high quality of the data.

[0020] As a preferred solution of the charging station loss prediction method based on the LSTM deep learning method described in the present invention, the step 2 process specifically includes:

[0021] S201: Data smoothing: Use moving average or exponential smoothing methods to process the time series data of charging stations to reduce the interference of short-term drastic fluctuations on model prediction:

[0022] S202: Outlier detection and correction: Detect equipment failure or abnormal data through the isolation forest algorithm, and correct the outliers by deletion, interpolation or mean substitution;

[0023] S203: Integrate expert knowledge features: Introduce key energy consumption influencing factors identified by experts, such as equipment aging and ambient temperature, to improve the accuracy and interpretability of the prediction model.

[0024] As a preferred solution of the charging station loss prediction method based on the LSTM deep learning method described in the present invention, the specific process of step three is:

[0025] S301: LSTM model construction: Use a two-layer LSTM neural network, set 128 and 64 units, combine the fully connected layer with the ReLU activation function, and build the model;

[0026] S302: Robustness test: The robustness of the model to the prediction error of the input variable is tested by inserting 1% to 10% Gaussian noise to ensure that the model is stable under different input errors;

[0027] S303: PPO algorithm optimization: The state parameters in the LSTM model are used as reinforcement learning strategies, and these states are continuously adjusted through the PPO algorithm, so that it can optimize energy loss prediction in complex dynamic environments.

[0028] As a preferred solution of the charging station loss prediction method based on the LSTM deep learning method described in the present invention, the specific operation process of the PPO algorithm optimization is:

[0029] The model optimizes the policy network through multiple iterations and uses the reward function to gradually optimize the energy consumption prediction. It also uses a clipping mechanism to prevent the policy from updating too quickly, which may cause the model to become unstable.

[0030] Among them, the reward function is defined according to the prediction error, and the mean absolute error MAE is used as the measurement standard. The smaller the error, the higher the reward, thereby continuously optimizing the accuracy of energy loss prediction. The mean absolute error MAE is calculated as follows:

[0031]

[0032] in, is the predicted value, y τ is the true value of the grid loss at time τ. The mean absolute error is chosen because it is less sensitive to possible outliers.

[0033] As a preferred solution of the charging station loss prediction method based on the LSTM deep learning method described in the present invention, the structure of the LSTM neural network is set as follows: 2 LSTM layers, 128 units in the first layer, and 64 units in the second layer; the fully connected layer is used to convert the output of the LSTM layer into a predicted value; the ReLU activation function is used for the fully connected layer; the loss function mean square error is set, which is suitable for regression tasks, and the Adam optimizer is used; the model input includes historical data on charging station losses and predicted data for the next 1 to 2 hours.

[0034] As a preferred solution of the charging station loss prediction method based on the LSTM deep learning method described in the present invention, the specific process of step 4 is as follows:

[0035] S401: Randomly search for hyperparameters: sample hyperparameter combinations of the number of LSTM units, number of layers, learning rate, and batch size, and perform model training;

[0036] S402: Performance evaluation: Evaluate the model performance under different hyperparameter combinations by comparing the mean absolute error and mean square error scores, and further fine-tune the model using the Adam optimizer.

[0037] The present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the above-mentioned charging station loss prediction method based on the LSTM deep learning method are implemented.

[0038] And a computer-readable storage medium having a computer program stored thereon, characterized in that: when the computer program is executed by a processor, the steps of the above-mentioned charging station loss prediction method based on the LSTM deep learning method are implemented.

[0039] Beneficial effects of the present invention:

[0040] 1. The present invention greatly improves the accuracy of energy loss prediction in charging stations by introducing the LSTM (long short-term memory) deep learning model. Traditional prediction methods usually rely on simple statistical models or linear regression models, which have limited ability to capture the complex nonlinear relationships in the energy loss of charging stations, resulting in low prediction accuracy. In contrast, the LSTM model specializes in processing time series data and can capture the long-term dependence and complex dynamic trends of energy loss in charging stations. The model proposed in the present invention comprehensively considers data in multiple dimensions, including core factors such as power flow, power conversion efficiency, utilization rate of charging equipment, line loss and equipment maintenance.

[0041] 2. In order to further improve the prediction accuracy, the present invention innovatively incorporates expert knowledge into the feature engineering part to perform multi-dimensional loss feature identification. For example, by combining the input of power system experts, factors such as equipment aging and climate change are identified, thereby improving the performance of the model in dealing with complex nonlinear relationships. This fusion method of expert knowledge and data-driven not only improves the prediction accuracy of the model, but also enhances the interpretability of the model, making the source of the loss characteristics clearer and facilitating optimization and adjustment by operators. Ultimately, the LSTM model can automatically learn complex nonlinear relationships from high-dimensional data and achieve high-precision energy loss prediction. Through accurate predictions, charging stations can more effectively optimize energy management and reduce operating costs, while improving charging efficiency, improving user experience, and promoting the widespread promotion of electric vehicle charging infrastructure.

[0042] 3. Reinforcement learning optimization and convergence speed improvement. The long-term memory unit of LSTM contains a random state parameter, and traditional optimization methods are inefficient in capturing dynamic changes. In order to optimize this process, the present invention combines a reinforcement learning algorithm, especially the PPO (Proximal Policy Optimization) algorithm, and effectively optimizes the state parameters of the LSTM model by introducing a Markov decision process. The PPO algorithm can improve the convergence speed of the LSTM model and reduce the convergence instability of the model under high-dimensional complex data by updating the policy gradient. Through this reinforcement learning optimization mechanism, the model can adapt to the changing charging station environment more quickly and still maintain high-precision prediction capabilities under complex load conditions.

[0043] 4. Centralized training and decentralized prediction architecture. The present invention adopts a centralized training and decentralized prediction framework. By collecting data from different charging stations, the model can be centrally trained on a central server, and after the training is completed, the miniaturized prediction model can be distributed and deployed to each charging station to achieve localized energy loss prediction. This architecture not only improves the training efficiency of the model, but also enhances the flexibility and scalability of the prediction system, ensuring that the prediction results of each station can achieve high-precision predictions with minimal computing resources, effectively supporting the real-time operation optimization of charging stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0045] Figure 1 A schematic diagram of the flow structure of a charging station loss prediction method based on the LSTM deep learning method proposed in the present invention;

[0046] Figure 2 This is a schematic diagram of a power grid data collection framework in a charging station loss prediction method based on an LSTM deep learning method proposed in the present invention;

[0047] Figure 3 This is a network architecture diagram with an input feature sequence length of 72 in a charging station loss prediction method based on the LSTM deep learning method proposed in the present invention. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0050] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0051] Reference Figure 1-Figure 2 , as an embodiment of the present invention, provides a charging station loss prediction method based on LSTM deep learning method, the method comprising the following steps:

[0052] Step 1: Data collection, including but not limited to: real-time power load data, power grid topology, time series data, environmental factors, and equipment status;

[0053] Specifically, the comprehensiveness and accuracy of these data are crucial for subsequent line loss analysis. Among them, the real-time load data of the power grid, including current, voltage, power factor, frequency, etc., is the basis of line loss analysis. These data are collected and transmitted in real time through various intelligent sensing devices in the distribution network, including smart meters, remote terminal units (RTUs), smart transformers, etc. The accuracy and stability of these devices directly affect the data quality, thereby affecting the results of line loss analysis. Grid topology information, that is, the connection relationship and configuration between the various parts of the grid, is equally important for understanding the operating status of the grid and identifying line loss paths. These data are usually obtained from the geographic information system (GIS) and asset management system of the power grid.

[0054] The specific data selected for this patent are as follows.

[0055]

[0056]

[0057] A total of 54 variables were selected with sufficiently long availability and sufficient data quality. The selection of these variables was based on data availability in real-time operations. Next, the dataset was divided into training / validation / testing periods with lengths of 6 years / 1 year / 1 year, respectively. The model inputs were preprocessed by quantile transformation to generate features that follow a uniform distribution on the interval [0,1]. The target values ​​of the model outputs were scaled to have zero mean and unit variance.

[0058] Step 2: Feature engineering: preprocessing and feature extraction of data, smoothing fluctuations, processing abnormal data, extracting expert knowledge features, and enhancing the predictive ability and interpretability of the model;

[0059] Specifically, first, the time series data collected during the operation of the charging station is smoothed. The operating data of the charging station (such as current, voltage, power, etc.) may experience short-term drastic fluctuations, affecting the model's capture of long-term trends. Through methods such as moving average or exponential smoothing, the interference of these short-term fluctuations can be reduced, making the data more stable, which in turn helps the LSTM model identify the long-term dependence of the data.

[0060] Secondly, there may be outliers due to equipment failure or other abnormal reasons during the data collection process. Through outlier detection technology, the isolation forest algorithm is used to identify and process these outliers. For the detected abnormal data, choose to delete or use interpolation or mean substitution to correct it to ensure the accuracy and continuity of the input data.

[0061] Considering that the sources of energy loss are multifaceted, we introduced loss feature identification that incorporates expert knowledge. By working with power system experts, we gain a deep understanding of the key factors affecting energy loss in charging stations, such as equipment aging, ambient temperature changes, and load fluctuations during specific time periods. These factors may be ignored or not fully captured in traditional data-driven methods. Incorporating these expert-identified features into the model not only improves the accuracy of energy loss prediction, but also enhances the interpretability of the model.

[0062] Compared with the existing purely data-driven methods, this method can improve prediction accuracy: integrating expert knowledge can capture important features that are not discovered in traditional methods, enrich the input information of the model, and improve prediction accuracy. At the same time, it also enhances the interpretability of the model: by introducing features identified by experts, the prediction results of the model are easier to understand, making it easier for operators to optimize the operation and maintenance strategies of charging stations in a targeted manner.

[0063] Step 3: Combine the PPO algorithm to establish an LSTM model. The LSTM model is used to capture the long-term dependencies in the energy loss time series data of the charging station. Combined with the PPO algorithm, the state parameters of the model are optimized to improve the prediction accuracy and convergence speed.

[0064] Specifically, the present invention designs and constructs a model based on LSTM (Long Short-Term Memory) neural network to process the sequence data of charging station losses and capture the long-term dependencies in the time series. The purpose of this model is to predict the total grid loss of the charging station system, with the ability to run once an hour, and predict the loss of the next hour 2 hours in advance based on historical data. The structure of the LSTM neural network is set as follows: 2 LSTM layers, 128 units in the first layer, and 64 units in the second layer; the fully connected layer is used to convert the output of the LSTM layer into a predicted value; the ReLU activation function is used for the fully connected layer; the loss function mean square error (MSE) is set, which is suitable for regression tasks, and the Adam optimizer is used.

[0065] Among them, the model input includes historical data of charging station losses and forecast data for the next 1 to 2 hours. The input sequence data reflects the past operating status of the charging station and the impact of external variables on losses, including historical power flow data, load demand data, line loss data, charging pile utilization rate, and forecast data for the next 1 to 2 hours. The present invention tests the robustness of the proposed model to the prediction error of the input variables by inserting Gaussian noise with a variance of 1% to 10% in all prediction inputs. The results show that the model has strong robustness to the prediction error, and the performance of the model has not changed significantly. The final model calculation diagram is shown in the figure. Figure 1 The recurrent neural network (RNN) sequentially processes a sequence of 70 past observations (x t-69 ,x t-68 …,x t ) and forecasts of 54 variables one to two hours in advance as model input.

[0066] Reference Figure 3 , the network architecture with an input feature sequence length of 72. 54 variables are selected as model input, and RNN serializes the first 70 observations and the last 2 predictions. The results of the model output The grid loss forecast for the charging station 2 hours in advance.

[0067] At each processing step, the RNN updates its hidden states h, which are passed as input to the next iteration. In the final processing step, the hidden states are linearly combined to compute the model output, which hopefully approximates the predicted grid loss hours. The present invention uses the mean absolute error (MAE) as the loss function:

[0068] in, is the predicted value, y τis the true value of the grid loss at time τ. The mean absolute error is chosen because it is less sensitive to possible outliers.

[0069] In the long-term memory unit of the LSTM model, there is a key state parameter that determines the model's memory of historical information and its ability to predict future outputs. In the traditional LSTM training process, this state parameter is usually initialized to a random value and then optimized by algorithms such as gradient descent. However, due to the complex nonlinear dynamic changes in the energy loss process of the charging station, relying solely on gradient descent may not capture this complex relationship, resulting in a slow convergence of the model and possibly insufficient accuracy.

[0070] To overcome this challenge, the present invention combines the PPO algorithm in deep reinforcement learning to optimize the state parameters of the LSTM model. Specifically, PPO is a deep reinforcement learning algorithm based on policy gradient, which performs well in processing high-dimensional and complex state spaces, and is particularly suitable for the scenario of energy loss prediction of electric vehicle charging stations in the present invention.

[0071] The core idea of ​​the PPO algorithm is to continuously update the policy network so that the model can continuously adjust its behavior strategy according to environmental feedback to obtain better long-term returns. In the present invention, the state parameters in the LSTM model are regarded as the "strategy" that needs to be optimized, and the prediction error of energy loss is used as the "reward" signal in reinforcement learning. In this way, the PPO algorithm can optimize the state parameters of the LSTM model, so that the model can more accurately predict energy loss when facing different load conditions, meteorological changes and equipment status.

[0072] The specific PPO algorithm process is as follows:

[0073] The state parameters of the LSTM model are optimized by combining the PPO (Proximal Policy Optimization) algorithm, generating state parameters based on the historical data of the charging station, and adjusting these states through the policy network, using the PPO clipping mechanism to prevent model instability. The reward function is defined according to the prediction error, using the mean absolute error (MAE) as the measurement standard. The smaller the error, the higher the reward, thereby continuously optimizing the accuracy of energy loss prediction. Experience replay technology further supports learning from historical data to prevent the model from falling into local optimality. The optimization goal of PPO is to maximize the cumulative reward, improve the prediction accuracy and convergence speed. The training process gradually improves the performance and stability of the LSTM model through multiple iterations and soft update mechanisms, so that it can better adapt to complex dynamic environments and achieve accurate energy consumption prediction and management.

[0074]

[0075] By combining PPO deep reinforcement learning, the LSTM model of the present invention can better adapt to the complex dynamic changes of energy loss in charging stations. First, the convergence speed of the model is significantly accelerated, and the energy loss pattern of charging stations can be learned in a shorter time; second, the prediction accuracy of the model is improved, especially under complex load conditions and environmental changes, the model's prediction of energy loss is more accurate.

[0076] Step 4: Hyperparameter tuning: Use the random search method to tune the hyperparameters of the LSTM model and find the optimal parameter combination to improve model performance.

[0077] Specifically, in order to optimize the performance of the LSTM model, the present invention also uses a random search method to adjust the hyperparameters of the model. The purpose of hyperparameter tuning is to find the optimal parameter combination that can maximize the prediction accuracy of the model. First, multiple different combinations are sampled from a fixed set of predefined hyperparameter values, including key parameters such as the number of LSTM units, the number of layers, the learning rate, and the batch size. After each sampling, the model is trained to observe the model performance under different parameter combinations. By comparing the MAE (mean absolute error) and MSE (mean square error) scores, it can be found that different weights initialized may cause a 5-10% variation in model performance. In order to further optimize the model, the present invention uses a stochastic gradient descent method and uses the Adam optimizer with a learning rate of 0.001 to adjust each individual LSTM model. In this way, the range of the optimal hyperparameters is gradually narrowed, and finally a parameter set that can effectively improve the accuracy of the model is determined, thereby improving the accuracy of predicting the energy loss of the charging station.

[0078] Step 5: After training, deploy the model to the charging station management system to realize real-time application of energy consumption prediction.

[0079] Specifically, after completing the model training and evaluation, the model will be deployed to the management system of the charging station for real-time application. By combining with the monitoring system of the charging station, the model can continuously obtain the real-time operation data of the charging station and make dynamic predictions based on these data. The LSTM model, which runs once an hour, can timely predict the future energy consumption of the charging station, help operators adjust charging strategies according to the latest prediction results, optimize energy consumption management, and reduce energy loss. At the same time, the present invention adopts a centralized training and decentralized prediction architecture, and collects historical data from each charging station for centralized training on a central server. The optimized miniaturized model is then deployed to each charging station to achieve local real-time energy loss prediction, thereby improving the accuracy of the prediction and the flexibility of the system.

[0080] This embodiment also provides a computer device, which is applicable to a charging station loss prediction method based on the LSTM deep learning method, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement a charging station loss prediction method based on the LSTM deep learning method proposed in the above embodiment.

[0081] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0082] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements a charging station loss prediction method based on an LSTM deep learning method as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a disk or an optical disk.

[0083] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A charging station loss prediction method based on LSTM deep learning method, characterized in that: The method comprises the following steps: Step 1: Data collection, including but not limited to: real-time power load data, power grid topology, time series data, environmental factors, and equipment status; Step 2: Feature engineering: preprocessing and feature extraction of data, smoothing fluctuations, processing abnormal data, extracting expert knowledge features, and enhancing the predictive ability and interpretability of the model; Step 3: Combine the PPO algorithm to establish an LSTM model. The LSTM model is used to capture the long-term dependencies in the energy loss time series data of the charging station. Combined with the PPO algorithm, the state parameters of the model are optimized to improve the prediction accuracy and convergence speed. Step 4: Hyperparameter tuning: Use the random search method to tune the hyperparameters of the LSTM model and find the optimal parameter combination to improve model performance. Step 5: Model deployment and application, using a centralized training and decentralized prediction architecture, collects historical data from each charging station and conducts centralized training on a central server; the optimized miniaturized model is then deployed to each charging station to achieve local real-time energy loss prediction.

2. According to claim 1, a charging station loss prediction method based on LSTM deep learning method is characterized in that: The real-time power load data collection method is to collect real-time power load data such as current, voltage, power factor and frequency through smart meters, remote terminal units and smart transformers; The grid topology information acquisition method is: extracting the connection relationship and configuration of each part of the grid from the grid geographic information system and asset management system; The time series data selection method is as follows: data is selected from 54 variables such as historical power flow data, load demand, line loss, and charging pile utilization rate, and input in time series to ensure the long-term availability and high quality of the data.

3. According to claim 2, a charging station loss prediction method based on LSTM deep learning method is characterized in that: The step 2 process specifically includes: S201: Data smoothing: Use moving average or exponential smoothing methods to process the time series data of charging stations to reduce the interference of short-term drastic fluctuations on model prediction: S202: Outlier detection and correction: Detect equipment failure or abnormal data through the isolation forest algorithm, and correct the outliers by deletion, interpolation or mean substitution; S203: Integrate expert knowledge features: Introduce key energy consumption influencing factors identified by experts, such as equipment aging and ambient temperature, to improve the accuracy and interpretability of the prediction model.

4. According to claim 3, a charging station loss prediction method based on LSTM deep learning method is characterized in that: The specific process of step three is: S301: LSTM model construction: Use a two-layer LSTM neural network, set 128 and 64 units, combine the fully connected layer with the ReLU activation function, and build the model; S302: Robustness test: The robustness of the model to the prediction error of the input variable is tested by inserting 1% to 10% Gaussian noise to ensure that the model is stable under different input errors; S303: PPO algorithm optimization: The state parameters in the LSTM model are used as reinforcement learning strategies, and these states are continuously adjusted through the PPO algorithm, so that it can optimize energy loss prediction in complex dynamic environments.

5. According to claim 4, a charging station loss prediction method based on LSTM deep learning method is characterized in that: The specific operation process of the PPO algorithm optimization is as follows: The model optimizes the policy network through multiple iterations and uses the reward function to gradually optimize the energy consumption prediction. It also uses a clipping mechanism to prevent the policy from updating too quickly, which may cause the model to become unstable. Among them, the reward function is defined according to the prediction error, and the mean absolute error MAE is used as the measurement standard. The smaller the error, the higher the reward, thereby continuously optimizing the accuracy of energy loss prediction. The mean absolute error MAE is calculated as follows: in, is the predicted value, y τ is the true value of the grid loss at time τ. The mean absolute error is chosen because it is less sensitive to possible outliers.

6. A charging station loss prediction method based on LSTM deep learning method according to claim 5, characterized in that: The structure of the LSTM neural network is set as follows: there are 2 LSTM layers, 128 units in the first layer, and 64 units in the second layer; the fully connected layer is used to convert the output of the LSTM layer into a predicted value; the ReLU activation function is used for the fully connected layer; the loss function is set to mean square error, which is suitable for regression tasks, and the Adam optimizer is used; the model input includes historical data on charging station losses and predicted data for the next 1 to 2 hours.

7. A charging station loss prediction method based on LSTM deep learning method according to claim 6, characterized in that: The specific process of step 4 is as follows: S401: Randomly search for hyperparameters: sample hyperparameter combinations of the number of LSTM units, number of layers, learning rate, and batch size, and perform model training; S402: Performance evaluation: Evaluate the model performance under different hyperparameter combinations by comparing the mean absolute error and mean square error scores, and further fine-tune the model using the Adam optimizer.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a charging station loss prediction method based on the LSTM deep learning method as described in any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a charging station loss prediction method based on an LSTM deep learning method as described in any one of claims 1 to 7 are implemented.

Images