Electric vehicle charging platform selection method based on charging efficiency and smart grid

Abstract

The present invention provides a method for selecting an electric vehicle charging platform based on charging efficiency and smart grids, which relates to the field of planning and management technology. By constructing a digital twin model of the battery, the electrochemical states such as SOC, SOH and temperature gradient are synchronized in real time; deep Q learning is used to train the main charging strategy and the auxiliary charging strategy respectively, realizing multi-objective trade-offs under the battery degradation model bundle; in actual operation, the main / auxiliary strategies respectively generate the first and second scheduling feedbacks, providing a multi-dimensional evaluation basis for different platforms; the multi-source feedback is fused into a hybrid reward, and weighted ranking is performed in combination with the platform load and efficiency information, and finally the platform with the highest score is selected to dynamically adjust the output power characteristic vector and update the degradation coefficient online; closed-loop optimization from digital twins, reinforcement learning to multi-strategy fusion is realized, which can not only improve the charging efficiency in a highly volatile power grid environment, but also reduce the battery life loss rate.

Description

Technical Field

[0001] The present invention relates to the field of planning and management technology, and in particular to a method for selecting an electric vehicle charging platform based on charging efficiency and smart grids. Background Art

[0002] In large-scale charging platforms, multiple electric vehicles are connected to the smart grid for charging simultaneously. Existing solutions often focus solely on grid load balancing or charging efficiency, overlooking the profound impact of battery health (SOH) on overall scheduling. With the integration of large-scale distributed renewable energy, grid volatility has increased, making it difficult for traditional static scheduling to balance grid security with battery life optimization. In recent years, digital twin technology has emerged in the industrial and transportation sectors. By constructing multi-physics models of physical objects, it enables multi-scale simulation and online correction of equipment operating states. Reinforcement learning has also achieved breakthroughs in intelligent scheduling, enabling autonomous learning of optimal strategies based on high-dimensional state vectors.

[0003] Prior art, published under the publication number CN112070300A, is titled "A Method for Selecting Electric Vehicle Charging Platforms Based on Multi-Objective Optimization." This method uses the total distance from the electric vehicle's current location to the charging platform, the total time to complete a charge, and the total cost as sub-objective functions. It also uses the electric vehicle's maximum range, road congestion, and the user's acceptable total charging time as constraints, upon which multi-objective optimization is performed. The method includes: 1) obtaining a charging request from an electric vehicle; 2) analyzing the charging request, road congestion, and information about nearby charging platforms to establish a multi-objective optimization mathematical model; and 3) using a deep learning algorithm to determine the optimal charging platform that meets the user's needs.

[0004] Although existing technologies have achieved certain results in load balancing and charging efficiency, they still have significant deficiencies in overall scheduling optimization:

[0005] First, traditional scheduling often targets grid load or single charging efficiency, ignoring the profound impact of battery SOH on charging speed and lifespan. This results in high-efficiency charging strategies exacerbating battery degradation in long-term operation.

[0006] Secondly, static or empirical power allocation cannot adapt to the time-varying characteristics of grid volatility and vehicle charging demand, making it difficult to optimize efficiency while ensuring grid security.

[0007] Thirdly, in multi-platform and multi-vehicle concurrent scenarios, there is a lack of effective comprehensive evaluation indicators and dynamic selection mechanisms, making it impossible to sort and switch based on the comprehensive performance of different platforms.

[0008] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0009] The purpose of the present invention is to provide a method for selecting an electric vehicle charging platform based on charging efficiency and smart grid, so as to solve the problems raised in the above background technology.

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] The method for selecting an electric vehicle charging platform based on charging efficiency and smart grid includes the following steps:

[0012] Step S1: During a preset monitoring period, charging efficiency, smart grid load, and battery health data are collected for each electric vehicle connected to each charging platform. This data is used to build a digital twin model of the battery for each electric vehicle.

[0013] Generate the battery state parameter vector of each electric vehicle and the output power characteristic vector of the corresponding smart grid through the battery digital twin model;

[0014] Step S2: Using a deep Q-learning method, a primary charging strategy and a secondary charging strategy are trained based on the battery state parameter vector of each electric vehicle and the output power characteristic vector of the corresponding charging platform; the training process includes a battery degradation model;

[0015] Step S3: The trained primary charging strategy and secondary charging strategy are deployed on edge servers respectively and integrated with the smart grid dispatching system. During the actual charging process, the primary charging strategy and secondary charging strategy generate corresponding first dispatching feedback results and second dispatching feedback results for each charging platform respectively.

[0016] Step S4: The first scheduling feedback result and the second scheduling feedback result are integrated and analyzed to generate mixed rewards for each charging platform; these mixed rewards are used to sort the corresponding charging platforms and select the optimal charging platform. The selected optimal charging platform will be used to adjust the output power characteristic vector of each electric vehicle charging pile corresponding to the corresponding smart grid in real time, and dynamically update the degradation model.

[0017] Compared with the existing technology, the beneficial effects of the present invention are: by constructing a battery digital twin model, the electrochemical states such as SOC, SOH and temperature gradient are synchronized in real time; deep Q learning is used to train the main charging strategy and the auxiliary charging strategy respectively, realizing multi-objective trade-offs under the battery degradation model bundle; in actual operation, the main / auxiliary strategies generate the first and second scheduling feedback respectively, providing multi-dimensional evaluation basis for different platforms; finally, the multi-source feedback is fused into a hybrid reward, and weighted ranking is performed in combination with the platform load and efficiency information, and the platform with the highest score is finally selected to dynamically adjust the output power characteristic vector and update the degradation coefficient online; closed-loop optimization from digital twins, reinforcement learning to multi-strategy fusion is realized, which can not only improve the charging efficiency by more than 20% in a highly volatile power grid environment, but also reduce the battery life loss rate by 15%, and has the innovation and practicality of grid security, battery life extension and charging efficiency improvement. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the overall method of the present invention. DETAILED DESCRIPTION

[0019] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. 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.

[0021] Example 1:

[0022] See also Figure 1 , the present invention provides a technical solution:

[0023] The method for selecting an electric vehicle charging platform based on charging efficiency and smart grid includes the following steps:

[0024] Step S1: During a preset monitoring period, charging efficiency, smart grid load, and battery health data are collected for each electric vehicle connected to each charging platform. This data is used to build a digital twin model of the battery for each electric vehicle.

[0025] Generate the battery state parameter vector of each electric vehicle and the output power characteristic vector of the corresponding smart grid through the battery digital twin model;

[0026] Further explanation: 1.1) The charging efficiency of each charging platform j and the electric vehicle i on the corresponding charging platform is connected through the data preprocessing module , smart grid load and battery health data Perform unified sampling, missing interpolation, anomaly removal, and Min–Max normalization processing according to a fixed sampling period. The specific implementation steps are as follows:

[0027] Set the charging platform: ; where j represents the index mark of the charging platform, and M is the total number of charging platforms;

[0028] Set up electric vehicles on the platform: ; where i represents the index of the electric vehicle, represents the total number of electric vehicles on charging platform j;

[0029] In this embodiment, "unified sampling" is achieved through time alignment, and the sampling period is unified to 1s;

[0030] The Min–Max normalization is characterized as ;

[0031] in ; is the minimum / maximum value of historical observations; The corresponding electric vehicle i on charging platform j The observed value at time ; and .

[0032] 1.2) The battery status parameter calculation module is based on the normalized battery health data of each electric vehicle i on each charging platform j after preprocessing and charging efficiency Constructing battery state parameter vector ; The specific implementation steps are as follows:

[0033] Will It is defined as a four-dimensional vector with components:

[0034] ; ; ; ;

[0035] in, They represent the state of charge and health status of electric vehicle i on charging platform j, respectively, in units of %; represents the temperature gradient of electric vehicle i on charging platform j, in °C; and and The values ​​are limited to the interval (0,1); min and max are the index marks of the minimum and maximum values ​​respectively, the same below, no further description is given; is the battery state parameter vector Normalized charging efficiency component of electric vehicle i on charging platform j; is the normalized state of charge component of electric vehicle i on charging platform j; is the normalized health status component of electric vehicle i on charging platform j; is the normalized temperature gradient component of electric vehicle i on charging platform j;

[0036] 1.3) Output power characteristic calculation module based on the normalized smart grid load of each charging platform j after preprocessing , load change rate and charging efficiency of corresponding electric vehicle i Construct output power characteristic vector ; Specifically expressed as:

[0037] Will It is defined as a three-dimensional vector, and each component is expressed as:

[0038] ; ; ;

[0039] in, is the real-time smart grid load of charging platform j; is the historical minimum / maximum value of the smart grid load change rate; is the real-time smart grid load change rate of charging platform j; Indicates that the charging platform is Smart grid load at all times; is the normalized smart grid load component on charging platform j; is the normalized smart grid load change rate component on charging platform j; is the power characteristic vector Normalized charging efficiency component of electric vehicle i on charging platform j;

[0040] 1.4) The battery state parameter vector of each electric vehicle i on each charging platform j is calculated through the feature fusion module and the output power characteristic vector Fusion according to preset weights generates reinforcement learning input feature vector The specific implementation content is as follows: Set the fusion coefficient ,and ,but ;

[0041] This embodiment sets 、 ; Vector dimensions: , dim represents the dimension of the vector, that is, the number of elements contained in the vector; all components are in the interval (0,1); is the battery state parameter vector The weight of is the output power characteristic vector The weight of .

[0042] Step S2: Using a deep Q-learning method, a primary charging strategy and a secondary charging strategy are trained based on the battery state parameter vector of each electric vehicle and the output power characteristic vector of the corresponding charging platform; the training process includes a battery degradation model;

[0043] Further explanation: In a multi-charging platform and multi-electric vehicle charging network, it is necessary to calculate the number of charging platforms j and each electric vehicle i connected to it based on the obtained feature vectors. Training the optimal charging strategy; specifically including:

[0044] 2.1) Define the state space of each electric vehicle i on each charging platform j , discrete action space And a compound reward function based on charging efficiency and battery degradation , used to train the Q network model; the specific implementation content is:

[0045] The state space Represented as a 7-dimensional vector ;

[0046] Discretize the action space Characterized by ; In turn, they represent the proportion of charging power to rated power;

[0047] Battery degradation model including normalized degradation cost Specific characteristics are ;in is the normalized degradation cost; and the normalized degradation cost ;

[0048] Define the reward function as ;

[0049] in, is the battery state parameter vector Normalized charging efficiency component of electric vehicle i on charging platform j at time t; is the selected charging power ratio; is the degradation coefficient; and is the reward weight of the corresponding parameter.

[0050] 2.2) By building an online behavioral network , target network and Experience Replay Buffer , using a DQN structure with a sigmoid output layer to ensure stable training and normalized Q values ​​within the interval (0,1); the specific implementation content is as follows:

[0051] In this embodiment, the network structure of the Q network model is set as follows:

[0052] Input: 7-dimensional state vector, corresponding to ;

[0053] Hidden: two layers with 64 units each, ReLU activation; Output: 5-dimensional, activated by sigmoid, ;

[0054] In this implementation, the hyperparameters are set as ; The target network update is every Step synchronization; among them, is the learning rate; is the discount factor used to calculate the weight of future rewards; B is the batch size;

[0055] Calculate for each batch ; and the network weight parameters Perform gradient descent; where, and are the behavior / target network weights, respectively; is the new battery state parameter vector after the action; is the current battery state parameter vector, specifically represented as ; Indicates action, from ; Indicates the proportion of the selected charging power to the rated power; Indicates taking action in state s After , the estimated value of future cumulative discounted rewards; is the expected value of the sampled batch data; y is the target Q value; r1 is the immediate reward; is the post-update action; is the representation loss function.

[0056] 2.3) The original Q value of the network output Perform Min–Max normalization and define the state confidence based on the normalized Q value , used to evaluate the reliability of the main charging strategy selection under the current state; and provide a basis for the next step of extracting the main charging strategy. The specific implementation content is as follows:

[0057] The normalized Q value is represented as ; and need to ensure ;and An index label indicating "original";

[0058] Setting state confidence ; which sets .

[0059] in, : original output of the behavior network; is the normalized Q value; and They are the minimum / maximum Q values ​​statistically obtained in the experience replay buffer; is the state confidence.

[0060] 2.4) Setting the confidence threshold , when the state confidence satisfies When the deterministic main charging strategy is extracted in this state , used for daily quick decision-making; otherwise, keep backup strategies to avoid invalid decisions;

[0061] The specific implementation contents are as follows:

[0062] The confidence threshold is taken as ; Fine-tune based on platform experience;

[0063] The main charging strategy is extracted as follows:

[0064]

[0065] in, is the main charging strategy; Yes The variable value with the maximum value; Dd represents the abbreviation of "to be determined";

[0066] 2.5) Based on state confidence Dynamically adjust Softmax temperature , and extract the randomized secondary charging strategy based on the full action set , used for balanced scheduling when load is sudden or uncertain;

[0067] The specific implementation contents are as follows:

[0068] In this embodiment, the Softmax temperature is set The minimum and maximum values ​​of , ; Dynamic temperature is calculated as ;make sure .

[0069] The price sub-charging strategy extraction representation is ;

[0070] in, is the dynamic temperature; is the secondary charging strategy distribution.

[0071] 2.6) Comparison through consistency detection module and The optimal action is to use the main charging strategy if the two are consistent, otherwise randomly switch between the two with a fixed probability p to balance efficiency and robustness. The specific implementation content is as follows:

[0072] Consistency check for ;examine ;

[0073] If they are consistent, the final strategy If they are inconsistent, the main charging strategy is selected with probability p=0.5, otherwise the randomized secondary charging strategy is selected. ,Right now ;in, is the randomized secondary charging strategy The optimal action; p is the fusion probability, and in this embodiment, p is set to 0.5; is the ultimate online strategy; It means "probability is".

[0074] 2.7) Combining state confidence, platform load fluctuations, and policy drift detection, adaptively adjust the online deployment cycle, experience replay capacity, and fine-tune trigger conditions. The specific implementation steps are as follows:

[0075] Adaptive deployment cycle ;

[0076] in, ; ; ; is the standard deviation of the platform grid load, which represents the load fluctuation; is the output value of the adaptive deployment cycle; and are the minimum and maximum values ​​of the deployment period, respectively; is the load fluctuation sensitivity coefficient, which represents the mapping ratio of the platform grid load fluctuation to the online deployment cycle adjustment amount;

[0077] Calculate the playback buffer capacity after dynamic adjustment ;in; ; is the buffer expansion factor, which measures the sensitivity of reward fluctuations to buffer capacity adjustments. Represents the compound reward function variance;

[0078] when When it increases, it means that the strategy is unstable under different sampling conditions. In order to retain more historical samples for smooth training, Will increase linearly;

[0079] when When it approaches 0, it means that the strategy has converged well. Approximately equal to the base capacity ;

[0080] coefficient The larger it is, the more sensitive the system is to reward fluctuations and the larger the buffer adjustment range is; conversely, the more conservative it is.

[0081] The policy drift detection trigger fine-tuning is as follows:

[0082] Calculate the main charging strategy distribution Compared with the baseline one hour ago KL divergence of ;

[0083] like Batch fine-tuning is started immediately; batch fine-tuning is part of the training process in step 2.2).

[0084] This embodiment not only achieves the precise extraction and fusion of the primary and secondary charging strategies, but also introduces adaptive cycles, dynamic buffering, drift detection, and federal collaboration in online deployment and fine-tuning, significantly improving the adaptability, stability, and global performance of the strategy.

[0085] Step S3: The trained primary charging strategy and secondary charging strategy are deployed on edge servers respectively and integrated with the smart grid dispatching system. During the actual charging process, the primary charging strategy and secondary charging strategy generate corresponding first dispatching feedback results and second dispatching feedback results for each charging platform respectively.

[0086] Further explanation: 3.1) The trained main charging strategy and secondary charging strategy It is encapsulated as a lightweight container image and deployed on each edge server, providing low-latency inference services through a RESTful interface. The specific implementation content is as follows:

[0087] Export the policy model to ONNX format, size .

[0088] And build the Docker image: ;

[0089] The inference engine in this embodiment uses ONNX-Runtime, and the batch size , ensuring single request latency .in, is the exported model size; : base image size; : inference batch size; is the single inference latency.

[0090] 3.2) At each charging decision, the main charging strategy interface and the secondary charging strategy interface are called respectively, and their output is directly used as the first scheduling feedback result Feedback results from the second dispatch Specific implementation includes:

[0091] First scheduling feedback result Specifically: ; This embodiment ensures .

[0092] Second scheduling feedback result Specifically ; is the secondary charging strategy distribution;

[0093] 3.3) Feedback on the first scheduling result Feedback results from the second dispatch Fusion is performed by fixed weights, and conflict detection is performed on multiple vehicle scheduling suggestions through thresholds to generate fusion feedback. ; This embodiment integrates feedback The weighted fusion representation of ;

[0094] in, , used to ensure ; Set the fusion feedback threshold ; If there are multiple electric vehicles on the same charging platform j and and and , then the suboptimal feedback is adjusted downward and the weight is reset to ; ;in, is the fusion weight; is the fusion feedback threshold; is the conflict down-regulation coefficient; Indicates that all optional actions Find the maximum value; This means that after summing up the recommended charging power ratios for all electric vehicles on the same charging platform, if the total exceeds 1.0, it means that the combined scheduling recommendations for all vehicles have exceeded the maximum available power of the platform.

[0095] 3.4) Fusion Feedback Sent to the smart grid dispatching system, and the final dispatching instructions are calculated based on the global load bundle and delay compensation mechanism .

[0096] The global load bundle in this embodiment is ;in ;

[0097] Delay compensation is network delay , the predicted state advance ,in is the rate of change of the battery state parameter vector; The specific characteristics are ; Final state input , updated feedback and instructions. is the final battery state parameter vector;

[0098] Publish payload using MQTT protocol ;in, is the rated power of charging platform j; is the maximum allowed power of charging platform j.

[0099] Step S4: The first scheduling feedback result and the second scheduling feedback result are integrated and analyzed to generate mixed rewards for each charging platform; these mixed rewards are used to sort the corresponding charging platforms and select the optimal charging platform. The selected optimal charging platform will be used to adjust the output power characteristic vector of each electric vehicle charging pile corresponding to the corresponding smart grid in real time, and dynamically update the degradation model.

[0100] Further explanation: 4.1) Based on the integrated feedback of all electric vehicles on each charging platform and normalized Q value , calculate the expected feedback value for electric vehicle i, and take the average normalization of the vehicle expected value to generate the mixed reward of charging platform j ;

[0101] In this embodiment, the vehicle expected feedback value is calculated as ; Need to ensure ;

[0102] Define the mixed reward of charging platform j The calculation formula is ;

[0103] in is the total number of electric vehicles connected to charging platform j; is the feedback distribution after fusion and conflict detection.

[0104] 4.2) Combination Calculate the load utilization of charging platform j , for mixed rewards Calculate weighted scores and sort in descending order to select the optimal platform with the highest score The specific implementation contents are as follows:

[0105] Define the load utilization of charging platform j for ;in is the maximum number of vehicles that can be connected to the charging platform j; Perform weighted score calculation to obtain ;

[0106] The optimal platform selection is characterized as ;in, is the weighted score of charging platform j; It is the best platform.

[0107] 4.3) Output power characteristic vector Perform linear combination with fixed weights to generate the dynamic output ratio of the charging pile corresponding to electric vehicle i in charging platform j , and accordingly set the adjusted final output power characteristic vector; the specific implementation content includes:

[0108] The weight selection rules of this embodiment are: ; ; ; ; Calculate the dynamic output ratio of the charging pile corresponding to electric vehicle i: ;

[0109] The selected optimal charging platform will be used to adjust the output power characteristic vector of each electric vehicle charging pile in the corresponding smart grid in real time and dynamically update the degradation model, specifically:

[0110] Set the output power characteristic vector as ;in is the rated output vector of charging platform j;

[0111] in, , , is the linear combination weight of the corresponding parameters; is the final output power characteristic vector.

[0112] 4.4) Battery capacity change after each charge Corresponding dynamic output ratio , calculate the instantaneous degradation coefficient, and based on the instantaneous degradation coefficient Degradation coefficient to be updated Exponential smoothing is used for online updating; the specific implementation is as follows:

[0113] In this embodiment, the instantaneous degradation coefficient is set to ; The degradation coefficient to be updated Characterized by ;

[0114] Exponential smoothing is updated to ; All subsequent degradation calculations are based on the updated ;

[0115] in, It is the change in battery capacity after a single charge; and are the old and new values ​​of the degradation coefficient, respectively; is the smoothing coefficient.

[0116] Example 2:

[0117] Four charging platforms are deployed in a regional smart grid, which are respectively called "Platform A", "Platform B", "Platform C" and "Platform D". The number of electric vehicles connected to each platform is 12, 16, 10, and 14 vehicles respectively; the maximum number of vehicles that can be connected to each platform 20 vehicles in total. Historical load range Set to [200kW, 800kW], load change rate historical range Set to [–40kW, 40kW]. The linear combination weight is ; and ; Smoothing coefficient .

[0118] Distribution of fusion feedback from connected vehicles per platform and normalized Q value Perform calculations. In the simulation test, according to the formula Get the expected feedback of each vehicle, and take the arithmetic average of the expected feedback of vehicles in the platform to get the platform mixed reward ;

[0119] The average expected feedback values ​​of the four platforms in this test were 0.68, 0.60, 0.75, and 0.55, respectively.

[0120] For platform weighted score ranking and optimal selection: current utilization of each platform They are 0.60, 0.80, 0.50 and 0.70 respectively. The calculated scores were 0.272, 0.120, 0.375, and 0.165, respectively. Platform C scored the highest and was therefore selected as the optimal charging platform. .

[0121] The output power characteristic vector construction and dynamic output ratio calculation contents are as follows:

[0122] After "Platform C" is selected, the following measurements and calculations are performed on it and all vehicles on the platform:

[0123] Real-time smart grid load They are 450, 600, 300, and 700 respectively;

[0124] Load change rate They are 20, –10, 30, and –20 respectively;

[0125] Average normalized charging efficiency They are 0.60, 0.70, 0.80 and 0.65 respectively.

[0126] Get the three-dimensional characteristic components of each platform respectively, and then press Calculate dynamic output ratio In this test, the average dynamic output ratios of the platforms were 0.608, 0.596, 0.696, and 0.567, with the largest value corresponding to "Platform C".

[0127] Selected After that, the battery capacity change is measured after each charge is completed. Calculate the instantaneous degradation coefficient ;

[0128] In this test, the calculated 0.222, 0.209, 0.114, 0.277 respectively. Then use exponential smoothing After the update, the platform C The lowest, indicating that battery degradation is most effectively controlled under dynamic output adjustment.

[0129] This embodiment verifies the innovation and superiority of the present invention in platform optimization, output characteristic adaptation, and online updating of degradation models through real measurement data and simulation calculations.

[0130] Table 1 Study on the optimal charging platform and the benefits of real-time adjustment:

[0131]

[0132] The table data analysis is as follows:

[0133] The ranking and optimal selection of charging platforms can be regarded as a multi-criteria decision-making problem, the core of which is to use mixed rewards to and load utilization Weighted score To determine the optimal platform; then, based on the characteristic vector constructed by the platform state and the corresponding dynamic output ratio For real-time power regulation; ultimately The degradation coefficient is updated using a quadratic relationship as input, achieving online model adaptation. The following section analyzes the data from the four platforms in the table layer by layer, and quantitatively explains the interactive changes of each parameter and the technical interval division.

[0134] 5) Mixed Rewards and Weighted Score Interval Division:

[0135] 5.1) High score range Platform C: , exceeding the threshold of 0.30 by 25.0%. Prioritize selection to ensure a balance between "high performance and low load".

[0136] 5.2) Medium score range ;

[0137] The weighted score of Platform A is 0.272, exceeding the lower limit by 0.15, reaching 81.3%.

[0138] Platform D’s weighted score is 0.165, exceeding the lower bound by 0.15, reaching 10.0%.

[0139] It can serve as a backup or fault-tolerant platform, and scoring and load balancing need to be determined manually or by secondary strategies.

[0140] 5.3) Low score range ;

[0141] Platform B has a weighted score of 0.120; it will not be selected for the time being to reduce dependence on high-utilization, low-mix reward platforms.

[0142] like Increase by 10%, and If unchanged, A corresponding increase of 10%;

[0143] like Increase by 10%, and If unchanged, A corresponding reduction of 10%;

[0144] Therefore, this embodiment accurately controls the balance between "reward enhancement" and "load suppression" through a linear complementary method, and is interpretable and adjustable.

[0145] Finally, according to the formula Select platform C as the optimal platform.

[0146] 6) For the output characteristic vector and dynamic output ratio range:

[0147] 6.1) High output ratio range Platform C: , exceeding the threshold of 0.65 by 7.1%. Allocate higher power output and shorten charging time.

[0148] 6.2) Medium output ratio range ; Platform A and B They are 0.608 and 0.596 respectively; maintain standard output rate and balance efficiency and equipment life.

[0149] 6.3) Low output ratio range ;

[0150] Platform Ding It is 0.567; increase output with caution and give priority to equipment protection.

[0151] The weight is the largest, so every 10% increase in charging efficiency can make Increase by 5%;

[0152] Secondly, the fluctuation of load change rate has an impact on It has a sensitivity of 3% / 10%×100%;

[0153] 7) Degradation coefficient online update interval:

[0154] 7.1) Deterioration range ;

[0155] All four platforms are within this range, with Platform C being the lowest. Even in high-output mode, degradation remains low, extending battery life by 3.5%.

[0156] Moderate degradation range ; No platform falls, improving the average lifespan consistency of the entire network.

[0157] When it increases by 10%, Increased by 21%, resulting in The decrease is nearly 17%, reflecting the secondary effect of charging power regulation on degradation control;

[0158] The use of an exponential smoothing factor of 0.9 can be found in Keep the model stable during mutations, with new values ​​accounting for 10% to ensure the robustness of online updates.

[0159] Multi-indicator fusion ranking: For the first time, fusion feedback is linearly coupled with platform utilization to achieve adjustable ranking of the dual bundles of "performance" and "resource occupancy".

[0160] Three-dimensional characteristic vector drive output: The characteristic vector is constructed by normalizing the load, load change rate and efficiency, and a dynamic output ratio is generated according to the adjustable weight, taking into account both real-time response and system stability.

[0161] Degradation model quadratic effect update: using The quadratic relationship realizes load-degradation coupling and is updated online with exponential smoothing, effectively balancing short-term performance and long-term life.

[0162] Example 3:

[0163] The following experimental table compares a traditional static scheduling strategy with the present invention's "Electric Vehicle Charging Platform Selection Method Based on Charging Efficiency and Smart Grid" in a highly fluctuating power grid environment. Through six sets of simulation tests with different fluctuation amplitudes, it is verified that the present invention improves charging efficiency by more than 20% while reducing battery life loss by 15%.

[0164] Table 2 Study on charging efficiency:

[0165]

[0166] Fluctuation amplitude: Characterizes the load variation range of the smart grid during charging.

[0167] Charging efficiency: the ratio of unit charging energy to input energy, taking the average value of each group of tests.

[0168] Life loss rate: The ratio of battery capacity attenuation due to degradation after daily charging, taking the daily average value.

[0169] The efficiency improvement and life loss reduction are both: (strategy of the present invention - traditional strategy) / traditional strategy × 100%.

[0170] As can be seen from the data in the table, under extreme conditions where the fluctuation range extends from ±20% to ±70%, the strategy of the present invention consistently achieves a charging efficiency improvement of more than 20% (minimum 20.0%, maximum 20.8%) compared with traditional scheduling, and the battery life loss rate is reduced by an average of about 17.0% compared with the traditional strategy (range 15.0%–20.0%), which fully demonstrates the significant beneficial effects of the method of the present invention in a highly volatile power grid environment.

[0171] It should be noted that: All calculation formulas in this application document use regression analysis including but not limited to machine learning algorithms to deeply analyze the relevant parameters collected and identify their natural trends and relationships. Use professional software, such as Python's Scikit-learn library or R language, to automatically generate mathematical models that match the data. Then, objectively evaluate the performance of the model through methods such as cross-validation, and combine continuous feedback and optimization to ensure that the created formula truly reflects the inherent laws of the data, thereby ensuring its effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula are dimensionally non-dimensionalized within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless technical means include but are not limited to Min-Max-Normalization and Z-Score standardization;

[0172] The technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random-access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0173] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0174] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. 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 method for selecting an electric vehicle charging platform based on charging efficiency and smart grid, characterized in that: The specific steps include: Step S1: During a preset monitoring period, data is collected from each electric vehicle connected to each charging platform; a digital twin model of the battery for each electric vehicle is constructed; and through the digital twin model, a battery state parameter vector and an output power characteristic vector of the corresponding smart grid are generated for each electric vehicle. The battery state parameter vector is composed of the state of charge, state of health, temperature gradient, and charging efficiency, and the output power characteristic vector is composed of the load, load change rate, and charging efficiency of the electric vehicle. The battery state parameter vector and the output power characteristic vector are fused according to preset weights to generate an input feature vector for the Q network model. Step S2: Define a state space, a discrete action space, and a composite reward function based on charging efficiency and battery degradation cost for training the Q network model; normalize the raw Q value output by the network, and calculate the state confidence based on the normalized Q value; when the state confidence is greater than or equal to a preset confidence threshold, extract the primary charging strategy based on the normalized Q value; dynamically adjust the Softmax temperature based on the state confidence, and use this temperature to extract the secondary charging strategy on the entire action set; compare the optimal actions of the primary and secondary charging strategies; if the two are consistent, adopt the primary charging strategy; otherwise, randomly switch between the two with a fixed probability p; Step S3: At each charging decision, the primary charging strategy and the secondary charging strategy generate corresponding first scheduling feedback results and second scheduling feedback results for each charging platform respectively; and generate fusion feedback based on the first scheduling feedback results and the second scheduling feedback results; Step S4: Based on the fused feedback and normalized Q value, the expected feedback value is calculated for each electric vehicle to generate a mixed reward for each charging platform; the load utilization of each charging platform is calculated, the mixed reward is weighted and scored, and the platform with the highest score is selected in descending order of score; the selected optimal charging platform will be used to adjust the output power characteristic vector of each electric vehicle charging pile corresponding to the corresponding smart grid in real time, and update the battery degradation cost.

2. The method for selecting an electric vehicle charging platform based on charging efficiency and smart grid according to claim 1, characterized in that: The charging efficiency of each charging platform j and the electric vehicle i on the corresponding charging platform is connected through the data preprocessing module , smart grid load and battery health data Perform unified sampling, missing interpolation, anomaly removal and Min–Max normalization according to a fixed sampling period; The battery status parameter calculation module is based on the normalized battery health data of each electric vehicle i on each charging platform j after preprocessing and charging efficiency Constructing battery state parameter vector ; The specific implementation steps are as follows: Will It is defined as a four-dimensional vector with components: ； ； ； ； in, They represent the state of charge and health status of electric vehicle i on charging platform j, respectively, in units of %; represents the temperature gradient of electric vehicle i on charging platform j, in °C; and and The values ​​are limited to the interval (0,1); min and max are the index marks of the minimum and maximum values ​​respectively; is the battery state parameter vector Normalized charging efficiency component of electric vehicle i on charging platform j; is the normalized state of charge component of electric vehicle i on charging platform j; is the normalized health status component of electric vehicle i on charging platform j; is the normalized temperature gradient component of electric vehicle i on charging platform j; The output power characteristic calculation module is based on the normalized smart grid load of each charging platform j after preprocessing , load change rate and charging efficiency of corresponding electric vehicle i Construct output power characteristic vector ; Specifically expressed as: Will It is defined as a three-dimensional vector, and each component is expressed as: ； ； ； in, is the real-time smart grid load of charging platform j; is the historical minimum / maximum value of the smart grid load change rate; is the real-time smart grid load change rate of charging platform j; Indicates that the charging platform is Smart grid load at all times; is the normalized smart grid load component on charging platform j; is the normalized smart grid load change rate component on charging platform j; is the power characteristic vector Normalized charging efficiency component of electric vehicle i on charging platform j; The battery state parameter vector of each electric vehicle i on each charging platform j is calculated by the feature fusion module and the output power characteristic vector According to the preset weight fusion, the input feature vector of the Q network model is generated .

3. The method for selecting an electric vehicle charging platform based on charging efficiency and smart grid according to claim 2, characterized in that: Define the state space of each electric vehicle i on each charging platform j , discrete action space And a compound reward function based on charging efficiency and battery degradation , used to train the Q network model; The state space Represented as a 7-dimensional vector ; Discretize the action space Characterized by , which represent the proportion of charging power to rated power respectively; the battery degradation model includes the normalized degradation cost of electric vehicle i on charging platform j ; The specific characteristics are ;in is the normalized degradation cost; and the normalized degradation cost ; Define the reward function as ; in, is the battery state parameter vector Normalized charging efficiency component of electric vehicle i on charging platform j at time t; is the selected charging power ratio; is the degradation coefficient; and is the reward weight of the corresponding parameter; By building an online behavioral network , target network and Experience Replay Buffer , using a DQN structure with a sigmoid output layer to ensure stable training and normalized Q values ​​within the interval (0,1); is the current battery state parameter vector; and are the behavior / target network weights, respectively; Indicates action, from .

4. The method for selecting an electric vehicle charging platform based on charging efficiency and smart grid according to claim 3, characterized in that: The original Q value of the network output Perform Min–Max normalization and define the state confidence of electric vehicle i on charging platform j based on the normalized Q value , used to evaluate the reliability of the main charging strategy selection under the current state; Setting confidence thresholds , when the state confidence satisfies When the deterministic main charging strategy is extracted in this state , used for daily quick decision-making; otherwise, keep backup strategies to avoid invalid decisions; Based on state confidence Dynamically adjust Softmax temperature , and extract the randomized secondary charging strategy based on the full action set , used for balanced scheduling during load bursts; Comparison through consistency detection module and If the two are consistent, the main charging strategy is adopted; otherwise, random switching is performed between the two with a fixed probability. Combining state confidence, platform load fluctuations, and policy drift detection, it adaptively adjusts the online deployment cycle, experience replay capacity, and fine-tunes trigger conditions.

5. The method for selecting an electric vehicle charging platform based on charging efficiency and smart grid according to claim 4, characterized in that: The main charging strategy completed by training and secondary charging strategy Encapsulated as a lightweight container image and deployed on each edge server, providing low-latency inference services through a RESTful interface; At each charging decision, the main charging strategy interface and the secondary charging strategy interface are called respectively, and their output is directly used as the first scheduling feedback result of electric vehicle i on charging platform j. Feedback results from the second dispatch ; Feedback results for the first dispatch Feedback results from the second dispatch Fusion is performed by fixed weights, and conflict detection is performed on multiple vehicle scheduling suggestions through thresholds to generate fusion feedback. ; Integrating feedback Sent to the smart grid dispatching system, and the final dispatching instructions are calculated based on the global load bundle and delay compensation mechanism ; in, The global load bundle is ;in ; Delay compensation is network delay , the predicted state advance ,in is the rate of change of the battery state parameter vector; The specific characteristics are ; Final state input , update feedback and instructions; is the final battery state parameter vector; Publish payload using MQTT protocol ;in, is the rated power of charging platform j; is the maximum allowed power of charging platform j.

6. The method for selecting an electric vehicle charging platform based on charging efficiency and smart grid according to claim 5, characterized in that: Based on the integrated feedback of all electric vehicles on each charging platform and normalized Q value , calculate the expected feedback value for electric vehicle i, and take the average normalization of the vehicle expected value to generate the mixed reward of charging platform j ; Combine Calculate the load utilization of charging platform j , for mixed rewards Calculate weighted scores and sort in descending order to select the optimal platform with the highest score .

7. The method for selecting an electric vehicle charging platform based on charging efficiency and smart grid according to claim 6, characterized in that: The output power characteristic vector Perform linear combination with fixed weights to generate the dynamic output ratio of the charging pile corresponding to electric vehicle i in charging platform j , and accordingly set the adjusted final output power characteristic vector.

8. The method for selecting an electric vehicle charging platform based on charging efficiency and smart grid according to claim 7, characterized in that: Based on the change in battery capacity after each charge Corresponding dynamic output ratio , calculate the instantaneous degradation coefficient , and based on the instantaneous degradation coefficient The degradation coefficient to be updated is updated online using the exponential smoothing method.

Images