Charging pile intelligent distribution system and charging pile based on dynamic load balancing

Abstract

The present invention discloses an intelligent distribution system for charging piles based on dynamic load balancing and a charging pile, and relates to the technical field of charging piles. The distribution module is used to receive the operation feature group of each charging pile at a first response time, generate a feature weight set, and simulate the importance judgment logic in the feature weight set according to the feature priority to realize dynamic load distribution of the charging pile; the process module is used to receive each source data in real time during the dynamic load distribution process, and perform multi-source analysis based on each source data and at a second response time to check the suddenness of charging pile failures, analyze communication interruption, predict charging demand and power supply, and construct a comprehensive risk index in combination with evidence theory; the update module is used to generate a judgment signal according to the value of the comprehensive risk index, and the judgment signal includes an update instruction and an iteration result.

Description

Technical Field

[0001] The present invention relates to the technical field of charging piles, and in particular to a charging pile intelligent distribution system and charging piles based on dynamic load balancing. Background Art

[0002] As the number of electric vehicles continues to grow, the distribution density and utilization efficiency of charging piles directly impact the user charging experience and the sustainable development of the industry. Consequently, smart charging pile systems have emerged. These systems leverage the Internet of Things, big data, and artificial intelligence technologies to enable remote monitoring, fault warnings, and resource scheduling of charging piles. However, in actual operations, the dynamic load distribution of charging piles has become increasingly problematic. Charging demand varies significantly across regions and time periods, and some charging piles may become overloaded or idle, resulting in wasted resources and reduced charging efficiency. Therefore, how to achieve even distribution of dynamic loads among charging piles through scientific algorithms and improve overall service efficiency has become a core issue that needs to be addressed in the smart charging pile field.

[0003] Currently, traditional charging pile allocation methods are mostly static, making them difficult to adapt to unexpected situations during operation. For example, fixed priority sorting based on the location and power of charging piles can easily lead to load imbalance. For example, when large-scale centralized charging demands arise, traditional systems struggle to respond quickly and dynamically adjust allocation plans. This can cause some charging piles to frequently fail due to overload, while others are underutilized.

[0004] The limitations of traditional load distribution methods have directly triggered a series of chain reactions. From the user's perspective, problems such as long charging queues and charging interruptions reduce user satisfaction and even hinder the popularization of electric vehicles. From the operator's perspective, low charging station utilization not only results in idle resources and waste, but also increases operating costs due to frequent fault repairs. At the risk management level, a single risk assessment method makes it difficult for the system to cope with complex and changing operating environments. In the event of power shortages or network communication interruptions, load distribution cannot be adjusted in a timely manner, which may lead to regional charging service paralysis. These problems seriously restrict the efficient operation of charging station infrastructure, and innovative technologies are urgently needed to achieve dynamic load balancing and intelligent risk management. Summary of the Invention

[0005] In view of the deficiencies in the prior art, the present invention provides a charging pile intelligent distribution system and charging piles based on dynamic load balancing, which solve the problems in the above-mentioned background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a charging pile intelligent distribution system based on dynamic load balancing, comprising:

[0007] The allocation module is used to receive the operating feature group of each charging pile at the first response time, generate a feature weight set, and simulate the importance judgment logic in the feature weight set based on the feature priority to achieve dynamic load distribution of the charging piles;

[0008] The process module is used to receive data from various sources in real time during the dynamic load distribution process. Based on these data, it performs multi-source analysis at the second response time to check the suddenness of charging pile failures, analyze communication interruptions, predict charging demand and power supply, and build a comprehensive risk indicator based on evidence theory.

[0009] The update module is used to generate a determination signal according to the value of the comprehensive risk indicator, and the determination signal includes an update instruction and an iteration result.

[0010] Preferably, the allocation module includes a first receiving unit, a weight generating unit and an importance unit;

[0011] Receiving unit No. 1 is used to pre-set the response interval and, at the first response moment, receive the operating feature group of each charging pile through several groups of sensors. The operating feature group includes but is not limited to voltage, current, power, number of connections, charging speed, fault duration, operating duration and number of charging guns; and pre-process the operating feature group to eliminate duplicate data and fill missing values to obtain the pre-processed operating feature group.

[0012] Preferably, the weight generating unit is used to extract the maximum value of each type of feature in the pre-processed operating feature group, and combine it with the proportional relationship with the equipment health value of the corresponding charging pile to define the weights of each type of feature after processing; wherein, the equipment health value represents the ratio of the non-fault time and the operating time of the corresponding charging pile in the historical period.

[0013] Preferably, the importance unit is used to use the Pearson correlation coefficient algorithm to analyze the correlation between various features and the equipment health value, so as to obtain the Pearson correlation coefficients between various features and the equipment health value respectively, and then compare the size relationship between the Pearson correlation coefficients in combination with the Saaty1-9 scaling method to construct a judgment matrix;

[0014] Based on the judgment matrix, the power method is used to calculate the maximum eigenvalue and the corresponding eigenvector of the judgment matrix, and the eigenvector corresponding to the maximum eigenvalue is normalized to obtain the weight of the corresponding feature in each Pearson correlation coefficient. The weights of the corresponding features in each Pearson correlation coefficient are arranged in ascending order to determine the priority of each feature in the operation feature group. According to the weights of each feature defined in the priority and weight generation unit, combined with the hierarchical fusion method, the final comprehensive weight of the corresponding charging pile is calculated. According to the final comprehensive weight of the corresponding charging pile, the dynamic load distribution of the charging pile is performed, specifically:

[0015] The allocation priority index of each charging pile is calculated by comparing the number of connections of the corresponding charging pile with its corresponding final comprehensive weight. When there is a vehicle to be charged, the allocation priority index of each charging pile in the charging station is compared, and the charging pile corresponding to the smallest allocation priority index is selected as the preferred charging pile for the vehicle to be charged this time.

[0016] Preferably, the process module includes a second receiving unit and a multi-source analysis unit;

[0017] The second receiving unit is used to receive various source data in real time based on the importance unit. The source data includes the charging pile operation feature group, network communication feature group and power supply feature group; among them, the network communication feature group includes various features during network communication; the power supply feature group includes the real-time remaining capacity of the power grid, the real-time power demand of the charging pile and the regional power allocation upper limit.

[0018] Preferably, the multi-source analysis unit is used to construct a fault prediction model using a random forest algorithm according to the operation feature group of the charging pile at the second response moment, divide the operation feature group of the charging pile into a training set and a test set, use the training set to train the fault prediction model, adjust the parameters of the model to optimize the performance of the model, and then use the test set to verify the trained fault prediction model, evaluate the accuracy and recall rate of the model, input the operation feature group of the charging pile monitored in real time into the trained fault prediction model, so that it learns the mode of fault occurrence, and the fault prediction model outputs the probability of each charging pile failing in the future; according to the network communication feature group, a hidden Markov model is used to predict the network status, and the network status is divided into different level states, level states The states include normal, unstable and interrupted, and the signal strength range corresponding to each level of state is determined. The received network communication feature group is then used as the observation sequence, and the network state is used as the hidden state. The hidden Markov model is trained through the known observation sequence and the corresponding hidden state to learn the state transition probability and observation probability to predict the probability of future network interruption; based on the operating feature group of the charging pile, historical charging demand data is extracted, and the time series prediction algorithm is used to predict the charging demand in future time periods. Combined with cluster analysis, high-demand areas and time periods are divided to calculate the risk probability; based on the power supply feature group, the real-time remaining capacity of the power grid, the real-time power demand of the charging pile and the regional power allocation upper limit are extracted, and the supply risk probability is obtained through ratio calculation.

[0019] Preferably, the multi-source analysis unit is also used to take the probability of failure of each charging pile in the future period of time, the probability of future network interruption, the risk probability and the supply risk probability as independent sources of evidence based on evidence theory. For each independent source of evidence, the basic probability distribution is determined according to its own characteristics and data distribution to obtain the basic probability distribution function of the corresponding independent source of evidence, and each basic probability distribution function is synthesized multiple times according to the DS synthesis rule to calculate and obtain a comprehensive risk index.

[0020] Preferably, the update module includes a determination unit and an update unit;

[0021] The determination unit is configured to extract the final comprehensive weight of the corresponding charging pile in the importance unit as the weight at the second response moment when the comprehensive risk index does not exceed a preset risk threshold; otherwise, generate an update instruction.

[0022] Preferably, the updating unit is used to determine the update instruction, and based on the final comprehensive weight of the corresponding charging pile in the importance unit, adjust the final comprehensive weight of the corresponding charging pile in the importance unit by a Sigmoid function to update the final comprehensive weight of the corresponding charging pile;

[0023] The iterative result returns the updated weight of each charging pile to the importance unit to recalculate the allocation priority index of each charging pile and re-obtain the preferred charging pile for the vehicle to be charged next time;

[0024] According to different response times, the allocation priority index of each charging pile within the corresponding response time is generated in sequence, and the allocation priority index of each charging pile within the corresponding response time is sorted in time sequence to obtain the iterative results of different response times.

[0025] A charging pile, comprising a main control module, at least one detection module and at least one power supply module arranged in a cabinet;

[0026] The detection module is used to detect the operation feature group of the charging pile and send the operation feature group to the main control module;

[0027] The main control module is used to generate a corresponding determination signal based on the operation feature group and send the determination signal to the power supply module;

[0028] The power supply module is used to supply power to the charging pile and to distribute the dynamic load evenly based on the judgment signal.

[0029] Beneficial effects:

[0030] Regarding dynamic load distribution, the allocation module receives operational characteristics of each charging station, uses a scientific algorithm to generate feature weights, and simulates importance judgment logic based on feature priorities. This module fully considers multiple factors, such as voltage, current, and power, overcoming the limitations of traditional static or empirical allocation methods. It achieves dynamic and precise allocation of charging station resources, effectively avoiding situations where some charging stations are overloaded while others are idle, improving overall charging station efficiency, reducing user wait time, and optimizing the charging experience. During the load distribution process, the process module collects multi-source data in real time and integrates techniques such as random forest algorithms and hidden Markov models to comprehensively analyze risks such as charging station failures, communication interruptions, charging demand, and power supply. Integrating evidence theory to construct a comprehensive risk index, the system can proactively identify potential risks, enhancing risk warning and response capabilities compared to traditional single-risk assessment models. The update module flexibly generates decision signals based on the values of the comprehensive risk index. When the risk exceeds the threshold, it promptly updates the charging station weights and dynamically adjusts the load distribution strategy, ensuring the system maintains efficient and stable operation in complex and changing operating environments.

[0031] By applying the Pearson correlation coefficient algorithm, we quantitatively analyze the correlation between various features and device health values, obtain the Pearson correlation coefficient, and objectively measure the impact of features on device health based on data. This avoids the bias of subjective judgments and re-quantifies the initially calculated feature weights, shifting from vertical single-category data analysis to horizontal multi-category data analysis. We also construct a judgment matrix using the Saaty 1–9 scaling method, transforming abstract feature importance comparisons into calculable numerical relationships. This matrix systematically presents the relative importance of each feature, making complex multi-factor evaluations more logical and standardized. Compared to traditional methods that simply allocate resources based on a single indicator, this method effectively avoids uneven use of charging piles, improves overall resource utilization efficiency, ensures the stable and efficient operation of the charging pile system, and provides users with a higher-quality and more convenient charging service experience.

[0032] The determination unit in the update module works closely with the update unit to dynamically adjust the charging pile weights based on a comparison of the comprehensive risk index with a preset threshold. When the comprehensive risk index remains within the threshold, the original weights are maintained to ensure stable system operation. If the comprehensive risk index exceeds the threshold, an update command is immediately triggered. Based on the comprehensive risk index value, the update unit uses a formula to perform nonlinear adjustments to the charging pile weights. This allows the weight adjustment to respond quickly to high-risk situations while remaining relatively stable when the risk is low, ensuring both scientific and flexible weight adjustments. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a module diagram of the charging pile intelligent distribution system based on dynamic load balancing of the present invention. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0035] Example 1

[0036] See also Figure 1 The present invention provides a charging pile intelligent distribution system based on dynamic load balancing, comprising:

[0037] The allocation module is used to receive the operating feature group of each charging pile at the first response time, generate a feature weight set, and simulate the importance judgment logic in the feature weight set based on the feature priority to achieve dynamic load distribution of the charging piles;

[0038] The process module receives data from various sources in real time during dynamic load distribution. Based on this data, it performs multi-source analysis at the second response time to check the suddenness of charging pile failures, analyze communication interruptions, predict charging demand and power supply, and construct a comprehensive risk indicator based on evidence theory (DS theory).

[0039] By monitoring risks in real time and adjusting weights, charging piles with potential faults or risks can be avoided in a timely manner, reducing charging interruptions caused by emergencies and improving the overall reliability of the system.

[0040] The update module is used to generate a determination signal according to the value of the comprehensive risk indicator, and the determination signal includes an update instruction and an iteration result.

[0041] In this embodiment of the present invention, the allocation module receives a set of charging pile operational characteristics and uses scientific calculations such as the Pearson correlation coefficient and the Saaty 1-9 scaling method to determine the weights and priorities of each characteristic, thereby simulating human decision-making logic to achieve dynamic load allocation. For example, during peak hours, the module can prioritize charging tasks to charging piles with low load and high performance based on characteristics such as the number of charging pile connections, power, and equipment health, thereby reducing the situation where some charging piles are overloaded and some are idle.

[0042] The process module collects multi-source data in real time during the load distribution process, and uses advanced technologies such as random forest algorithms and hidden Markov models to analyze charging pile failures, network communications, charging demand and power supply respectively, and uses evidence theory to construct comprehensive risk indicators. For example, when a large-scale event in a certain area causes a surge in charging demand, the module can quickly predict demand changes and, combined with the power supply situation of the power grid, accurately assess the risks of each charging pile.

[0043] The update module generates a decision signal based on the comprehensive risk index. When the comprehensive risk index exceeds the threshold, it promptly issues an update command and updates the charging pile weights through weight adjustment, achieving dynamic weight optimization. For example, if the health of a charging pile device deteriorates and the risk of failure increases, the value of its comprehensive risk index increases. The update module will reduce the weight of the charging pile, reduce the allocation of new tasks, and ensure the stable operation of the entire system.

[0044] Example 2

[0045] Please refer to Figure 1 ,Specifically: the allocation module includes a No. 1 receiving unit, a weight generating unit and an ,importance unit;

[0046] Receiving unit No. 1 is used to pre-set the response interval and, at the first response moment, receive the operating feature group of each charging pile through several groups of sensors. The operating feature group includes but is not limited to voltage, current, power, number of connections, charging speed, fault duration, operating duration and number of charging guns; and pre-process the operating feature group to eliminate duplicate data and fill missing values to obtain the pre-processed operating feature group.

[0047] The weight generation unit is used to extract the maximum value of each type of feature in the pre-processed operating feature group and combine it with the proportional relationship between it and the equipment health value of the corresponding charging pile to define the weight of each type of feature after processing. Specifically, the ratio of each type of feature in the corresponding charging pile to the feature with the largest value among all its types of features is calculated, and the calculation result is the weight of each type of feature after processing. For better explanation, the algorithm is as follows: ;in, is the feature of the i-th charging pile The feature weights of is the feature of the i-th charging pile Number, is the maximum value of feature z among all charging piles, and n is the total number of charging piles;

[0048] Extract the maximum value of each feature in the pre-processed operation feature group and combine it with the proportional relationship between it and the equipment health value of the corresponding charging pile, specifically:

[0049] The purpose of calculating the ratio of various features to the equipment health value of the corresponding charging pile in advance is to align the subsequent weight acquisition with the equipment health value of the corresponding charging pile, which is equivalent to the reference feature.

[0050] The above calculation formula shows that the corresponding characteristics in the numerator are directly proportional to the device health value; if they are inversely proportional, the contents in the numerator and denominator in the calculation formula will be swapped.

[0051] Among them, the equipment health value H represents the ratio of the non-fault time and the operating time of the corresponding charging pile in the historical period.

[0052] In this embodiment, the allocation module and its subunits in the present invention achieve accurate quantification of charging pile operation characteristics and scientific allocation of resources through systematic data processing and weight calculation, effectively improving the refinement of charging pile management and resource utilization efficiency.

[0053] Receiver Unit 1, acting as a front-end sentinel for data collection, proactively collects operational characteristic data from each charging station at the first response moment, using pre-set response intervals. This data covers multiple dimensions, including voltage, current, and power. It then pre-processes the data to remove duplicates and fill in missing values. For example, in a cluster of charging stations, Receiver Unit 1 promptly discovered duplicate current data from a particular charging station due to a sensor anomaly. After processing, the data was verified to be accurate, providing a reliable foundation for subsequent analysis.

[0054] The weight generation unit acts like an intelligent analyst, extracting the maximum values of various features based on the operational feature data processed by the first receiving unit. These values are then normalized with the device health value to determine the weights of each feature. For example, if a charging station's historical fault-free time accounts for 90% of its operating time, indicating good health, the weight of features directly proportional to this factor (such as charging speed) will be increased in the weight calculation. For example, if a charging station has a fast charging speed and a high device health value, its charging speed feature weight is calculated to be 0.8. Meanwhile, if another charging station has a low device health value due to frequent faults, its charging speed feature weight may be only 0.3, even with the same charging speed. This differentiated weighting allows the system to prioritize charging tasks for charging stations with better performance and more stable status, avoiding resource misallocation, improving overall service quality and device utilization, and ensuring dynamic and efficient optimization of charging station resources in complex and changing operating environments.

[0055] Example 3

[0056] Please refer to Figure 1 Specifically: In order to make the calculation of various feature weights more logical and understandable, a calculation method based on priority sorting and normalization is used to convert each individual weight into a comprehensive weight;

[0057] The importance unit is used to analyze the correlation between various features and the device health value using the Pearson correlation coefficient algorithm to obtain the Pearson correlation coefficient between each feature and the device health value. In combination with the Saaty1-9 scaling method, the magnitude relationship between each Pearson correlation coefficient is compared to construct a judgment matrix. Each element in the judgment matrix reflects the relative ratio between two Pearson correlation coefficients.

[0058] Furthermore, for better explanation, the following is the expression of the judgment matrix A:

[0059] ;

[0060] Where m is the number of comparisons, which is equal to the total number of features. is the relative importance ratio of the first Pearson correlation coefficient to the second Pearson correlation coefficient; Example: If =3, then =1 / 3;

[0061] The equipment health value is measured by the ratio of historical fault-free time to operating time, which directly reflects the operating stability of the charging pile.

[0062] Based on the judgment matrix, the power method is used to calculate the maximum eigenvalue and corresponding eigenvector of the judgment matrix. The eigenvector corresponding to the maximum eigenvalue is normalized to obtain the weight of the corresponding feature in each Pearson correlation coefficient. The weights of the corresponding features in each Pearson correlation coefficient r are arranged in ascending order to determine the priority of each feature in the operation feature group. Based on the weights of each feature defined in the priority and weight generation unit, combined with the hierarchical fusion method, the final comprehensive weight of the corresponding charging pile is calculated;

[0063] In a judgment matrix, the maximum eigenvalue is the eigenvalue with the largest absolute value among the eigenvalues of the matrix;

[0064] The power method is an iterative method for calculating the principal eigenvalue (i.e., the eigenvalue with the largest absolute value) of a matrix and its corresponding eigenvector. The basic idea is to obtain a vector sequence by continuously multiplying an initial vector by the matrix on the left. As the number of iterations increases, the vector sequence will gradually converge to the principal eigenvector of the matrix. At the same time, the principal eigenvalue can be obtained through certain calculations.

[0065] Furthermore, based on the judgment matrix, the power method is used to calculate the maximum eigenvalue and corresponding eigenvector of the judgment matrix, specifically including:

[0066] Initialization: Select a non-zero initial vector x0, you can choose a vector of all 1s, and set the number of iterations u=0;

[0067] Iterative calculation: Calculation , find the element with the largest absolute value in y(u+1) and calculate the vector for the next iteration;

[0068] Convergence judgment: To judge whether the convergence condition is met, the absolute value of the difference between the vectors obtained from two adjacent iterations can be used to see whether it is less than a preset threshold. If the convergence condition is met, the iteration is stopped.

[0069] Calculate the maximum eigenvalue: When the iteration converges, the corresponding element is the maximum eigenvalue of matrix A.

[0070] Through hierarchical fusion, the importance differences of various weight factors are reflected, and each individual weight is mapped to the same value range (0-1), eliminating the influence of the original numerical values of different weights, making each weight comparable and fusible.

[0071] According to the final comprehensive weight of the corresponding charging pile, the dynamic load distribution of the charging pile is carried out as follows:

[0072] The allocation priority index of each charging pile is calculated by calculating the ratio of the number of connections of the corresponding charging pile to its corresponding final comprehensive weight;

[0073] Furthermore, for better explanation, it is reflected by the formula: ;in, is the allocation priority index of the i-th charging pile, is the number of connections of the i-th charging pile, is the final comprehensive weight of the i-th charging pile;

[0074] When there is a vehicle to be charged, the allocation priority indexes of the charging piles in the charging station are compared, and the charging pile corresponding to the smallest allocation priority index is selected as the preferred charging pile for the vehicle to be charged this time.

[0075] Among them, the corresponding eigenvector is the vector corresponding to the maximum eigenvalue, and its components represent the relative importance weight of each Pearson correlation coefficient in the whole;

[0076] The correlation between performance indicators was calculated using the Pearson correlation coefficient.

[0077] In this embodiment, the importance unit uses the Pearson correlation coefficient algorithm to deeply analyze the correlation between various operating characteristics of the charging pile and the equipment health value, and quantifies the Pearson correlation coefficient, thereby objectively reflecting the degree of influence of each characteristic on the equipment health. For example, if the Pearson correlation coefficient between the current characteristic and the equipment health value is close to 1, it indicates that the current size is highly positively correlated with the equipment health status, which needs to be considered in load distribution. Combined with the Saaty1-9 scaling method, the Pearson correlation coefficients are compared to construct a judgment matrix, and then the power method is used to calculate the maximum eigenvalue and eigenvector, and the normalization process is performed to obtain the weight of each feature, and then the feature priority is determined. Based on this, combined with the feature weight defined by the weight generation unit, the final comprehensive weight of the charging pile is calculated by the hierarchical fusion method.

[0078] During actual allocation, the importance unit calculates an allocation priority index, correlating the number of connected charging piles with the final overall weight, to numerically present the load allocation priority of each charging pile. For example, at a charging station, if one charging pile has a small number of connections but a high overall weight, its calculated allocation priority index will be small, while another charging pile has a large number of connections but a low overall weight, resulting in a large allocation priority index. The system will then prioritize assigning vehicles to the first charging pile to avoid overloading the second one and achieve load balancing.

[0079] During this process, the No. 1 receiving unit is responsible for collecting and preprocessing the charging pile operation data; the weight generation unit calculates the feature weight of the data; and the importance unit further determines the feature priority and comprehensive weight. The three work together to form a complete link from data processing to strategy execution for the dynamic load distribution of charging piles, effectively solving the problem of resource imbalance that is easily caused by traditional distribution methods.

[0080] Example 4

[0081] Please refer to Figure 1 ,Specifically: the process module includes the No. 2 receiving unit and the multi-source analysis unit;

[0082] The second receiving unit is used to receive various source data in real time based on the importance unit. The source data includes the charging pile operation feature group, network communication feature group and power supply feature group; among them, the network communication feature group includes various features during network communication; the power supply feature group includes the real-time remaining capacity of the power grid, the real-time power demand of the charging pile and the regional power allocation upper limit.

[0083] The multi-source analysis unit is used to construct a fault prediction model based on the operating feature group of the charging pile using a random forest algorithm at the second response time. The operating feature group of the charging pile is divided into a training set and a test set. The fault prediction model is trained using the training set and the model parameters are adjusted to optimize the model performance. The trained fault prediction model is then verified using the test set to evaluate the model's accuracy and recall rate. The operating feature group of the charging pile monitored in real time is input into the trained fault prediction model so that it learns the pattern of fault occurrence. The fault prediction model outputs the probability of each charging pile failing within a certain period of time in the future.

[0084] Based on the network communication feature set, a hidden Markov model is used to predict the network status. The network status is divided into different levels, including normal, unstable, and interrupted. The signal strength range corresponding to each level is determined. The received network communication feature set is then used as the observation sequence and the network status as the hidden state. The hidden Markov model is trained using the known observation sequence and the corresponding hidden state to learn the state transition probability and observation probability. The forward-backward algorithm is used to calculate the probability distribution of the hidden state under given model parameters and observation sequence, thereby obtaining the probability of future network interruption.

[0085] The commonly used training algorithm is the Baum-Welch algorithm, which is an iterative algorithm based on maximum likelihood estimation and is used to estimate the parameters of the hidden Markov model.

[0086] Based on the operating characteristics of the charging piles, historical charging demand data is extracted. A time series prediction algorithm (such as the Prophet algorithm) is used to predict charging demand in future time periods. Combined with cluster analysis, high-demand areas and time periods are divided to calculate the risk probability. Specifically, the predicted charging demand is subtracted from the maximum carrying capacity of the corresponding charging station. The numerical state of the calculated result is then compared with the maximum value of the predicted demand in all areas to ultimately obtain the risk probability.

[0087] Furthermore, for better explanation, we will further demonstrate it through the algorithm: ;in, is the risk probability of the i-th charging pile facing charging demand, is the predicted charging demand in the area where the i-th charging pile is located at time t, is the maximum carrying capacity of the i-th charging pile, that is, the maximum load that the charging pile can provide charging services to vehicles at the same time; The maximum value of the predicted demand for all areas; this formula expresses the risk probability of large-scale concentrated charging demand by calculating the difference between the predicted demand and the maximum carrying capacity of the charging pile and normalizing it to the range of 0 to 1.

[0088] The logic of obtaining risk probability is: first calculate , obtaining the difference between the predicted demand and the maximum carrying capacity of charging pile i at time t. If the difference is greater than 0, it indicates that the predicted demand exceeds the maximum carrying capacity of the charging pile, posing an overload risk. In this case, the difference is used as the numerator. If the difference is less than or equal to 0, it indicates that the predicted demand is within the charging pile's carrying capacity and there is no overload risk. In this case, the numerator is 0. The numerator is then divided by the maximum predicted demand across all regions to normalize the risk probability to a range of 0 to 1. This facilitates comparison and analysis of the risk levels of different charging piles facing large-scale concentrated charging demand.

[0089] Cluster analysis: Divide charging areas into clusters based on geographic location and charging station density. For each cluster, analyze the distribution of historical charging demand to identify high-demand areas and time periods. For example, a K-means clustering algorithm can be used to divide the area into K clusters based on similarities in charging demand.

[0090] Based on the power supply feature group, the real-time remaining capacity of the grid Er, the real-time power demand of the charging pile Ed, and the upper limit of regional power allocation Ea are extracted. The supply risk probability EF is obtained through ratio calculation, which is specifically: ;in, is the supply risk probability of the i-th charging pile, is the real-time power demand of the i-th charging pile, that is, the actual power consumed by the charging pile at the current moment. The regional power allocation upper limit Ea refers to the maximum power capacity that can be allocated to the charging pile in the area where the corresponding charging pile is located; the real-time remaining capacity of the power grid Er is the remaining power capacity available for allocation at the current moment;

[0091] The logic of obtaining the supply risk probability is: first calculate , which is the difference between the sum of the real-time power demands of all charging piles and the upper limit of regional power allocation, to measure whether the total power demand of charging piles in the region exceeds the upper limit of available power. If the difference is greater than 0, it indicates insufficient power supply and a risk exists, and this difference is used as the numerator. If the difference is less than or equal to 0, it indicates that the power supply can meet demand and there is no risk, and the numerator is 0. Finally, the numerator is divided by the real-time remaining capacity of the power grid to normalize the supply risk probability to a range of 0 to 1. This yields the probability of insufficient power supply in the region where charging pile i is located, which is used to assess the severity of power supply tension in the region.

[0092] In the embodiment of the present invention, the second receiving unit is responsible for collecting multi-source data such as the charging pile operation feature group, the network communication feature group and the power supply feature group in real time, providing rich and timely data support for the system operation. It acts as the information collector of the system to ensure the comprehensiveness and real-time nature of the data.

[0093] The multi-source analysis unit acts as the system's intelligent brain, deeply mining the collected data. It uses the random forest algorithm to build a fault prediction model. Through training and validation, it optimizes model performance and accurately predicts the probability of future charging pile failures. For example, by analyzing characteristics such as voltage fluctuations and operating time, it can predict in advance that a charging pile is likely to fail within three days due to component aging. This allows operations and maintenance personnel to conduct repairs in advance and minimize the impact of sudden failures on users.

[0094] Using hidden Markov models to predict network status can accurately estimate the risk of network disruption. For example, during large-scale events, the probability of a nearby charging station network disruption due to a surge in users can be determined in advance, allowing network optimization or backup plans to be prepared in advance. Using a time series prediction algorithm combined with cluster analysis to predict charging demand can effectively identify high-demand areas and time periods, such as around tourist attractions during holidays, accurately predicting charging peaks and avoiding overcrowding at charging stations.

[0095] By calculating supply risk probabilities based on power supply feature groups, the system can monitor the balance of power supply and demand in real time. For example, if the remaining capacity of a regional power grid is insufficient, the system can promptly adjust the power distribution of charging piles to prevent power overload. Through comprehensive analysis of multi-source data, the system can fully understand the operational risks of charging piles, providing a scientific basis for subsequent dynamic load distribution and weight adjustments, effectively improving the stability and reliability of the charging pile system and ensuring the user's charging experience.

[0096] Example 5

[0097] Please refer to Figure 1 Specifically: For better explanation, the probability of failure of each charging pile in the future, the probability of network interruption, the risk probability and the supply risk probability are added with symbols F, G, DF and EF respectively;

[0098] The multi-source analysis unit is also used to use the probability F of each charging pile failing in the future, the probability G of future network interruption, the risk probability DF, and the supply risk probability EF as independent evidence sources based on evidence theory. For each independent evidence source, the basic probability distribution is determined based on its own characteristics and data distribution to obtain the basic probability distribution function Y of the corresponding independent evidence source. Each basic probability distribution function Y is then synthesized multiple times according to the DS synthesis rule to calculate the comprehensive risk index CRI, which is specifically: ;in, is the comprehensive risk index of the i-th charging pile; 、 、 and are the basic probability distribution functions of the corresponding independent evidence sources in the i-th charging pile, It is the synthesis operator in DS synthesis rules.

[0099] For each independent source of evidence, the basic probability distribution is determined based on its own characteristics and data distribution. For example, the probability F of a charging pile failure can be mapped to a basic probability distribution function based on the statistical distribution of historical failure data and the current operating status of the charging pile. Assuming that the failure risk is divided into three levels: low, medium, and high, the basic probability is allocated based on the probability that the value of the failure probability F falls into different levels. For example, if the failure probability F is in the range of (0-0.3), the probability of belonging to the low risk level is 0.7, the probability of belonging to the medium risk level is 0.2, and the probability of belonging to the high risk level is 0.1.

[0100] DS synthesis rule is a method for fusing the basic probability distribution of multiple evidence sources. It can synthesize the basic probability distribution functions of different evidence sources to obtain a comprehensive basic probability distribution function, thereby realizing the fusion of information. operator, 、 、 and Gradually synthesize and finally obtain the basic probability distribution function of the comprehensive risk index , thereby comprehensively considering multiple risk factors and integrating information from multiple independent evidence sources.

[0101] In the Naive Bayes algorithm, the posterior probability of each category can be calculated based on the training data to determine the value of the basic probability distribution function.

[0102] The update module includes a determination unit and an update unit;

[0103] The determination unit is configured to extract the final comprehensive weight of the corresponding charging pile in the importance unit as the weight at the second response moment when the comprehensive risk index does not exceed a preset risk threshold; otherwise, generate an update instruction.

[0104] An updating unit, configured to determine an update instruction and, based on the final comprehensive weight of the corresponding charging pile in the importance unit, adjust the final comprehensive weight of the corresponding charging pile in the importance unit by a Sigmoid function to update the final comprehensive weight of the corresponding charging pile;

[0105] Furthermore, for better explanation, the algorithm is Obtain and update the final comprehensive weight of the corresponding charging pile; is the updated weight of the i-th charging pile, is the comprehensive risk index of the i-th charging pile, is the slope control parameter, which is used to control the steepness of the curve. The larger k is, The steeper the function is at the inflection point, the more dramatic the weight change; is a position control parameter used to control the position of the inflection point, that is, the CRI value when the weight begins to change significantly. When the comprehensive risk index is larger, it indicates that the comprehensive risk of the charging pile is higher, and its weight needs to be reduced accordingly to reduce the possibility of assigning new tasks to the charging pile.

[0106] When the comprehensive risk index approaches 1, Approaching 1, Approaching 0 (weight is greatly reduced);

[0107] When the comprehensive risk index approaches 0, Approaching 0 Approaching the final comprehensive weight Wz (the weight remains basically unchanged).

[0108] By collecting historical data, including charging station usage, fault records, and user feedback under different comprehensive risk indicators, we analyze the relationship between the comprehensive risk indicators and actual business results. By observing the data distribution and changing trends, we identify the comprehensive risk indicator value at which the weight begins to change significantly, which serves as a reference for the position control parameter. To determine the slope control parameter, we can try different slope control parameter values and plot a curve showing the adjustment coefficient as it changes with the comprehensive risk indicator. We then observe which k value best aligns the curve with business expectations, specifically, matching the weight change with the actual risk situation.

[0109] The iterative result returns the updated weight of each charging pile to the importance unit to recalculate the allocation priority index of each charging pile and re-obtain the preferred charging pile for the vehicle to be charged next time;

[0110] According to different response times, the allocation priority index of each charging pile within the corresponding response time is generated in sequence, and the allocation priority index of each charging pile within the corresponding response time is sorted in time sequence to obtain the iterative results of different response times.

[0111] In this embodiment of the present invention, the multi-source analysis unit utilizes evidence theory (DS theory), treating the probability of charging pile failure, network interruption probability, charging demand risk probability, and power supply risk probability as independent sources of evidence. Using a basic probability distribution function and DS synthesis rules, this multi-dimensional risk information is integrated into a comprehensive risk indicator. This approach avoids the one-sidedness of single risk assessments and more comprehensively and accurately reflects the actual risk status of charging piles. For example, if a charging pile has a low probability of failure but a high probability of power supply risk, the comprehensive risk indicator will take both into account, avoiding underestimation of the potential risk of the charging pile.

[0112] The decision unit compares the comprehensive risk index with a preset threshold and determines whether to trigger a weight update. If the comprehensive risk index does not exceed the threshold, the original weights are retained; otherwise, an update process is initiated to ensure the system can respond promptly to changes in risk.

[0113] Based on the update instructions, the update unit dynamically adjusts the weights of charging piles using a formula. Slope and position control parameters flexibly adjust the sensitivity and inflection points of weight changes. For example, during peak charging hours, if a charging pile's comprehensive risk index increases due to insufficient power supply, the system uses the update unit to reduce its weight and reduce the number of new tasks assigned. As the risk decreases, the weight can gradually increase again, achieving dynamic balance.

[0114] In short, the multi-source analysis unit is responsible for integrating multi-dimensional risk information and generating comprehensive evaluation indicators; the judgment unit is used to judge the risk status and decide whether to start weight update; the update unit dynamically adjusts the charging pile weight according to risk changes, thereby realizing intelligent and dynamic distribution of charging pile loads and ensuring efficient and stable operation of the system.

[0115] All the aforementioned parameters and calculated parameters were normalized (such as Z-score normalization) to eliminate the influence of different original numerical values, making the data comparable and fusible.

[0116] Example 6

[0117] Please refer to Figure 1 , specifically: a charging pile, comprising a main control module, at least one detection module and at least one power supply module arranged in a cabinet;

[0118] The detection module is used to detect the operation feature group of the charging pile and send the operation feature group to the main control module;

[0119] The main control module is used to generate a corresponding determination signal based on the operation feature group and send the determination signal to the power supply module;

[0120] The power supply module is used to supply power to the charging pile and to distribute the dynamic load evenly based on the judgment signal.

[0121] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The intelligent charging pile allocation system based on dynamic load balancing is characterized by: include, The allocation module is used to receive the operating feature group of each charging pile at the first response time, generate a feature weight set, and simulate the importance judgment logic in the feature weight set based on the feature priority to achieve dynamic load distribution of the charging piles; The process module is used to receive data from various sources in real time during the dynamic load distribution process. Based on these data, it performs multi-source analysis at the second response time to check the suddenness of charging pile failures, analyze communication interruptions, predict charging demand and power supply, and build a comprehensive risk indicator based on evidence theory. An update module is used to generate a determination signal based on the value of the comprehensive risk indicator, and the determination signal includes an update instruction and an iteration result; The weight generation unit is used to extract the maximum value of each type of feature in the pre-processed operating feature group and combine it with the proportional relationship between the maximum value and the equipment health value of the corresponding charging pile to define the weight of each type of feature after processing. The equipment health value represents the ratio of the non-fault time to the operating time of the corresponding charging pile in the historical period. The importance unit is used to analyze the correlation between various features and the equipment health value using the Pearson correlation coefficient algorithm to obtain the Pearson correlation coefficient between each feature and the equipment health value. In combination with the Saaty1-9 scaling method, the size relationship between the Pearson correlation coefficients is compared to construct a judgment matrix. Based on the judgment matrix, the power method is used to calculate the maximum eigenvalue and the corresponding eigenvector of the judgment matrix, and the eigenvector corresponding to the maximum eigenvalue is normalized to obtain the weight of the corresponding feature in each Pearson correlation coefficient. The weights of the corresponding features in each Pearson correlation coefficient are arranged in ascending order to determine the priority of each feature in the operation feature group. According to the weights of each feature defined in the priority and weight generation unit, combined with the hierarchical fusion method, the final comprehensive weight of the corresponding charging pile is calculated. According to the final comprehensive weight of the corresponding charging pile, the dynamic load distribution of the charging pile is performed, specifically: The allocation priority index of each charging pile is calculated by comparing the number of connections of the corresponding charging pile with its corresponding final comprehensive weight. When there is a vehicle to be charged, the allocation priority index of each charging pile in the charging station is compared, and the charging pile corresponding to the smallest allocation priority index is selected as the preferred charging pile for the vehicle to be charged this time.

2. The charging pile intelligent allocation system based on dynamic load balancing according to claim 1 is characterized in that: The allocation module includes a No. 1 receiving unit, a weight generating unit, and an importance unit; The first receiving unit is used to pre-set the response interval and, at the first response time, receive the operating characteristic group of each charging pile through several groups of sensors. The operating characteristic group includes voltage, current, power, number of connections, charging speed, fault duration, operating duration, and number of charging guns; The running feature group is preprocessed to remove duplicate data and fill in missing values to obtain the preprocessed running feature group.

3. The charging pile intelligent allocation system based on dynamic load balancing according to claim 2 is characterized in that: The process module includes the No. 2 receiving unit and the multi-source analysis unit; The second receiving unit is used to receive various source data in real time based on the importance unit. The source data includes the charging pile operation feature group, network communication feature group and power supply feature group; among them, the network communication feature group includes various features during network communication; the power supply feature group includes the real-time remaining capacity of the power grid, the real-time power demand of the charging pile and the regional power allocation upper limit.

4. The charging pile intelligent allocation system based on dynamic load balancing according to claim 3 is characterized by: The multi-source analysis unit is used to build a fault prediction model using the random forest algorithm according to the operation feature group of the charging pile at the second response moment, divide the operation feature group of the charging pile into a training set and a test set, use the training set to train the fault prediction model, adjust the parameters of the model to optimize the performance of the model, and then use the test set to verify the trained fault prediction model, evaluate the accuracy and recall rate of the model, input the operation feature group of the charging pile monitored in real time into the trained fault prediction model, so that it learns the mode of fault occurrence, and the fault prediction model outputs the probability of each charging pile failing in the future; according to the network communication feature group, the hidden Markov model is used to predict the network status, and the network status is divided into different level states. Including normal, unstable and interrupted, and determine the signal strength range corresponding to each level state, then use the received network communication feature group as the observation sequence, and the network state as the hidden state. The hidden Markov model is trained through the known observation sequence and the corresponding hidden state to learn the state transition probability and observation probability to predict the probability of future network interruption; according to the operation feature group of the charging pile, the historical charging demand data is extracted, and the charging demand in the future time period is predicted using the time series prediction algorithm. Combined with cluster analysis, the high-demand areas and time periods are divided to calculate the risk probability; according to the power supply feature group, the real-time remaining capacity of the power grid, the real-time power demand of the charging pile and the regional power distribution upper limit are extracted, and the supply risk probability is obtained through ratio calculation.

5. The charging pile intelligent allocation system based on dynamic load balancing according to claim 4 is characterized in that: The multi-source analysis unit is also used to treat the probability of failure of each charging pile in the future, the probability of future network interruption, the risk probability and the supply risk probability as independent evidence sources based on evidence theory. For each independent evidence source, the basic probability distribution is determined according to its own characteristics and data distribution to obtain the basic probability distribution function of the corresponding independent evidence source, and each basic probability distribution function is synthesized multiple times according to the DS synthesis rule to calculate and obtain the comprehensive risk index.

6. The charging pile intelligent allocation system based on dynamic load balancing according to claim 5 is characterized by: The update module includes a determination unit and an update unit; The determination unit is configured to extract the final comprehensive weight of the corresponding charging pile in the importance unit as the weight at the second response moment when the comprehensive risk index does not exceed a preset risk threshold; otherwise, generate an update instruction.

7. The charging pile intelligent allocation system based on dynamic load balancing according to claim 6 is characterized in that: An updating unit, configured to determine an update instruction and, based on the final comprehensive weight of the corresponding charging pile in the importance unit, adjust the final comprehensive weight of the corresponding charging pile in the importance unit by a Sigmoid function to update the final comprehensive weight of the corresponding charging pile; The iterative result returns the updated weight of each charging pile to the importance unit to recalculate the allocation priority index of each charging pile and re-obtain the preferred charging pile for the vehicle to be charged next time; According to different response times, the allocation priority index of each charging pile within the corresponding response time is generated in sequence, and the allocation priority index of each charging pile within the corresponding response time is sorted in time sequence to obtain the iterative results of different response times.

8. A charging pile, using the system according to any one of claims 1 to 7, characterized in that: It includes a main control module, at least one detection module and at least one power supply module arranged in the cabinet; The detection module is used to detect the operation feature group of the charging pile and send the operation feature group to the main control module; The main control module is used to generate a corresponding determination signal based on the operation feature group and send the determination signal to the power supply module; The power supply module is used to supply power to the charging pile and to distribute the dynamic load evenly based on the judgment signal.

Images