Charging demand prediction system and method based on multi-source heterogeneous data fusion
By fusion of multi-source heterogeneous data, combined with deep learning and historical analysis, the low accuracy problem of electric vehicle charging demand forecasting is solved, and charging demand forecasting and load balancing with high accuracy and real-time response are achieved.
Patent Information
- Application Number
- CN202510779257.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-11
- Publication Date
- 2025-09-19
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing electric vehicle charging demand forecasting methods find it difficult to establish accurate probabilistic models, and machine learning-based methods fail to fully consider multi-source factors, resulting in low prediction accuracy.
The method of multi-source heterogeneous data fusion is adopted. Through data collection, preprocessing, deep learning, historical analysis and load assessment, combined with the impact of charging convenience, the graph attention network is used to predict charging demand, and load assessment and early warning are performed.
It improves the accuracy and reality relevance of charging demand forecasts, enhances the system's risk perception and response capabilities, and achieves load balancing and resource optimization.
Smart Images
Figure CN120675048A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of charging demand prediction, and in particular to a charging demand prediction system and method based on multi-source heterogeneous data fusion. Background Art
[0002] The transformation of the automotive industry to electrification has become an important trend in the sustainable development of society. Electric vehicles, with their good environmental protection and energy-saving advantages, have become an important direction for the future development of the automotive industry. Therefore, it can be seen that the scale of electric vehicles in my country will further expand in the future.
[0003] The main existing research methods for electric vehicle charging demand can be divided into two categories: traditional prediction methods based on probability models and intelligent prediction methods based on machine learning. The method based on probability models mainly establishes a probability model for the charging behavior of electric vehicles, and then uses the Monte Carlo simulation method to generate a set of random data to simulate the charging behavior of electric vehicle groups for prediction. Considering the randomness of the areas where electric vehicles are connected to the power grid, it is very difficult to establish an accurate probability model, and the above methods can only make simple estimates. At the same time, due to the lack of real charging data, most studies simulate the charging behavior of electric vehicles and may be far from the actual situation. Intelligent prediction methods based on machine learning mainly use machine learning algorithms and models to model from a data-driven perspective, and make predictions by analyzing and training historical data to obtain a prediction model. However, most studies use shallow models such as ANN, SVM, KNN, etc. to make one-dimensional time series predictions, and the model learning ability and expression ability of complex systems are limited;
[0004] With the further development of mobile charging technology and the increasing number of mobile charging piles, people are no longer just considering whether charging is possible, how to reduce grid impact, and how to lower charging costs when forecasting EV charging demand. They are also considering factors such as weather conditions and the distance between the charging pile and the car to be charged. This further affects the accuracy of EV charging demand forecasts.
[0005] Therefore, it is difficult to establish an accurate probability model for traditional electric vehicle charging demand forecasting methods, and they can only achieve a simple estimation of electric vehicle charging demand. The existing intelligent forecasting methods based on machine learning do not consider the impact of other multi-source factors on charging demand, resulting in low accuracy in electric vehicle charging demand forecasting. Summary of the Invention
[0006] In response to the shortcomings of the existing technology, the present invention provides a charging demand prediction system and method based on multi-source heterogeneous data fusion, which has the advantages of more accurately predicting the charging demand of electric vehicles and solves the above-mentioned technical problems.
[0007] To achieve the above-mentioned object, the present invention provides the following technical solution: a charging demand prediction system based on multi-source heterogeneous data fusion, comprising a data acquisition module, a preliminary prediction module, a charging demand prediction module and a load assessment module;
[0008] The data acquisition module is used to collect charging-related data in the current area and pre-process the charging-related data to obtain processed charging-related data;
[0009] The preliminary prediction module is equipped with a deep learning unit, and the processed charging-related data is input into the deep learning unit to obtain a preliminary prediction value CBYCZ;
[0010] The charging demand prediction module includes a historical charging demand analysis unit, a charging convenience evaluation unit, and a charging demand prediction unit. The historical charging demand analysis unit analyzes the historical charging demand of the current area to obtain a historical charging impact coefficient. The charging convenience evaluation unit is used to analyze the convenience of the current area to obtain a charging convenience impact coefficient. The charging demand prediction unit obtains a final charging prediction value ZZYCZ based on a comprehensive analysis of the preliminary prediction value, the historical charging impact coefficient, the charging convenience impact coefficient, and the historical charging demand of the current area.
[0011] The load evaluation module is used to evaluate the final charging prediction value, issue an early warning based on the charging prediction value, and execute an early warning strategy.
[0012] As a preferred technical solution of the present invention, the data acquisition module collects charging-related data of the current area, including temperature, humidity, precipitation probability, user charging time, SOC during charging, and charging power of the charging pile.
[0013] As a preferred technical solution of the present invention, the charging-related data is preprocessed using Z-score standardization, and the deviation of the preprocessed values from their mean is analyzed, and values beyond k standard deviations from the mean after preprocessing are screened out to obtain the processed charging-related data.
[0014] As a preferred technical solution of the present invention, the deep learning unit is specifically a graph attention network, and the input is processed charging-related data, the output is power, and the output power is used as a preliminary prediction value.
[0015] As a preferred technical solution of the present invention, the historical charging demand analysis unit analyzes the historical charging demand of the current area and obtains the specific expression of the historical charging impact coefficient as follows:
[0016]
[0017] Among them, LSCDYX represents the historical charging impact coefficient of the i-th evaluation, XQZ n Indicates the nth environment consistency coefficient. If the current area is consistent with the nth environment in the statistics of the actual charging demand for the i-th time, then XQZ n =1, otherwise XQZ n =0.
[0018] As a preferred technical solution of the present invention, the charging convenience evaluation unit is used to analyze the convenience of the current area, and obtain the specific expression of the charging convenience influence coefficient as follows:
[0019]
[0020] Among them, PCZ h It represents the h-th deviation value that affects the user's arrival at the current area. The higher the value, the worse the convenience. CDBJ represents the charging convenience impact coefficient.
[0021] As a preferred technical solution of the present invention, the current regional historical charging demand is obtained by reading the historical power demand database by the charging demand forecasting module, and the specific steps are as follows:
[0022] Read the historical power demand database based on LSCDYX i Filter one by one and select all LSCDYX i The top M actual charging demands in the ranking are averaged as the historical charging demand of the current region
[0023] As a preferred technical solution of the present invention, the charging demand forecasting unit obtains the final charging forecast value ZZYCZ based on a comprehensive analysis of the preliminary forecast value, the historical charging impact coefficient, the charging convenience impact coefficient, and the current area's historical charging demand. The specific expression is as follows:
[0024]
[0025] in, represents the historical charging demand in the current area, CBYCZ represents the preliminary forecast value, ZZYCZ represents the final charging forecast value, and CDBJ represents the charging convenience impact coefficient.
[0026] As a preferred technical solution of the present invention, the load assessment module is used to assess the charging prediction value, issue an early warning based on the charging prediction value, and execute the early warning strategy in the following specific steps:
[0027] When the final charging prediction value ZZYCZ exceeds the preset charging load, an early warning is issued, and the redundancy of adjacent charging areas is obtained, and scheduling is carried out by the adjacent areas.
[0028] The present invention also provides a charging demand prediction method based on multi-source heterogeneous data fusion, which includes the following steps:
[0029] Step 1: Collect charging-related data in the current area and pre-process the charging-related data to obtain processed charging-related data;
[0030] Step 2: Input the processed charging-related data into the deep learning unit to obtain a preliminary prediction value
[0031] Step 3: Analyze the historical charging demand in the current area and obtain the historical charging impact coefficient;
[0032] Step 4: Analyze the convenience of the current area and obtain the impact coefficient of charging convenience;
[0033] Step 5: Read the historical power demand database to obtain the current area's historical charging demand;
[0034] Step 6: Obtain the final charging forecast value based on a comprehensive analysis of the preliminary forecast value, the historical charging impact coefficient, the charging convenience impact coefficient, and the current area's historical charging demand;
[0035] Step 7: Evaluate the final charging prediction value, issue an early warning based on the charging prediction value, and implement the early warning strategy.
[0036] Compared with the existing technology, the present invention provides a charging demand prediction system and method based on multi-source heterogeneous data fusion, which has the following beneficial effects:
[0037] The present invention uses a closed-loop process from data collection and preprocessing to deep learning prediction, and then to multi-factor comprehensive analysis and load assessment. It not only utilizes the advantages of big data and deep learning, but also combines the influence of historical experience and actual charging convenience, ensuring the high accuracy and reality relevance of the prediction results. At the same time, the load assessment and early warning mechanism enables the system to have strong risk perception and response capabilities. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 Schematic diagram of the system framework of the present invention;
[0039] Figure 2 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0040] 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.
[0041] See also Figure 1-Figure 2 , a charging demand prediction system based on multi-source heterogeneous data fusion, including data acquisition module, preliminary prediction module, charging demand prediction module and load assessment module;
[0042] The data acquisition module is used to collect charging-related data in the current area and pre-process the charging-related data to obtain processed charging-related data;
[0043] The data acquisition module collects charging-related data in the current area, including temperature, humidity, precipitation probability, user charging time, SOC during charging, and charging power of charging piles. The charging-related data are preprocessed using Z-score standardization. At the same time, the deviation of the preprocessed values from their mean is analyzed, and values beyond k standard deviations from the mean after preprocessing are screened out to obtain the processed charging-related data. The module comprehensively collects multi-source heterogeneous charging-related data in the current area and preprocesses them to ensure the quality and consistency of the input data, providing high-quality basic data support for subsequent models.
[0044] The preliminary prediction module is equipped with a deep learning unit, and the processed charging-related data is input into the deep learning unit to obtain the preliminary prediction value CBYCZ;
[0045] The deep learning unit is specifically a graph attention network, with processed charging-related data as input and power as output, which is used as a preliminary prediction value. By being equipped with a deep learning unit, it can automatically learn complex nonlinear relationships and implicit features, and generate preliminary prediction values based on preprocessed data. It has strong generalization capabilities and the ability to adapt to dynamic changes;
[0046] The graph attention network includes an input layer, a GAT layer, and an output layer. The input layer node feature matrix is of shape (N, F), where N is the number of charging piles and F is the feature dimension, specifically temperature, humidity, precipitation probability, user charging time, SOC during charging, and charging power of the charging pile.
[0047] The expression of the GAT layer is as follows:
[0048]
[0049] Among them, ab represents different nodes, N(a) represents the node set, Represents the importance weight of node b to node a, that is, the attention coefficient, and the sum is 1, W l represents the weight matrix of the lth layer, represents the feature representation of node a at layer l+1, Indicates the feature representation of node b at layer l, B l represents the bias at layer l;
[0050] The output layer is used to convert the output into power for output. The specific expression is: MAXCHARGE i Indicates the maximum charging capacity (power), baseCHARGE i represents the base power, RELU represents the nonlinear activation function, and the loss function is the mean square error.
[0051] The charging demand forecasting module includes a historical charging demand analysis unit, a charging convenience assessment unit, and a charging demand forecasting unit. The historical charging demand analysis unit analyzes the historical charging demand of the current area to obtain a historical charging impact coefficient. The charging convenience assessment unit is used to analyze the convenience of the current area to obtain a charging convenience impact coefficient. The charging demand forecasting unit obtains a final charging forecast value ZZYCZ based on a comprehensive analysis of the preliminary forecast value, the historical charging impact coefficient, the charging convenience impact coefficient, and the historical charging demand of the current area. The module comprehensively considers multiple factors such as historical charging demand and charging convenience, and uses the historical charging impact coefficient and the convenience impact coefficient to assist in revising the preliminary forecast, thereby improving the accuracy and practicality of the forecast and enhancing the responsiveness to regional characteristics and user behavior.
[0052] The historical charging demand analysis unit analyzes the historical charging demand of the current area and obtains the specific expression of the historical charging impact coefficient as follows:
[0053]
[0054] Among them, LSCDYX represents the historical charging impact coefficient of the i-th evaluation, XQZ n Indicates the nth environment consistency coefficient. If the current area is consistent with the nth environment in the statistics of the actual charging demand for the i-th time, then XQZ n =1, otherwise XQZ n =0, which naturally has the ability to resist noise and exclude abnormal historical samples; the convenience evaluation weakens the extreme impact of a single deviation value through multi-factor averaging. The combination of the two can effectively improve the stability of the system in complex scenarios. The environment is consistent, including weather (rain or no rain), holidays, promotions and price reductions in a store in the region, and other similar situations for comparison.
[0055] The charging convenience evaluation unit is used to analyze the convenience of the current area. The specific expression of the charging convenience impact coefficient is as follows:
[0056]
[0057] Among them, PCZ h It represents the h-th deviation value that affects users' arrival at the current area. The higher the value, the worse the convenience. CDBJ represents the charging convenience impact coefficient, and the deviation value is a percentage, such as the density and distribution balance of charging piles in the area; the operating status of charging piles (idle / faulty / congested); traffic accessibility (peak / off-peak traffic conditions); user queue length and average waiting time; user evaluation data and behavioral preferences. Technical personnel can choose any of them as long as the higher the value, the worse the convenience.
[0058] The current regional historical charging demand is obtained by reading the historical power demand database through the charging demand forecast module. The specific steps are as follows:
[0059] Read the historical power demand database based on LSCDYX I Filter one by one and select all LSCDYX i The top M actual charging demands in the ranking are averaged as the historical charging demand of the current region
[0060] The charging demand forecasting unit obtains the final charging forecast value ZZYCZ based on a comprehensive analysis of the preliminary forecast value, the historical charging impact coefficient, the charging convenience impact coefficient, and the current regional historical charging demand. The specific expression is as follows:
[0061]
[0062] in, represents the current historical charging demand in the region, CBYCZ represents the preliminary forecast value, ZZYCZ represents the final charging forecast value, and CDBJ represents the charging convenience impact coefficient. It combines the preliminary forecast value (the forecast output of the model based on existing data) and the regional historical charging demand. This can make full use of historical experience and refer to the new forecast results to enhance the robustness of the model. Charging convenience has an adjustment effect on charging demand, increasing the demand in areas with high convenience and appropriately lowering the demand in areas with low convenience, which is more in line with actual user behavior and reflects real constraints.
[0063] The load assessment module is used to evaluate the final charging prediction value, issue early warnings based on the charging prediction value, and execute early warning strategies.
[0064] The load assessment module is used to evaluate the charging prediction value, issue warnings based on the charging prediction value, and implement the warning strategy. The specific steps are as follows:
[0065] When the final charging prediction value ZZYCZ exceeds the preset charging load, an early warning is issued, and the redundancy of adjacent charging areas is obtained, and scheduling is carried out by adjacent areas. In order to alleviate the load pressure, the module automatically identifies the relevant adjacent charging areas, queries the current load margin and remaining capacity (redundancy) of these areas, and evaluates whether the overload part can be shared. According to the available capacity of the adjacent area, the load assessment module calls the scheduling mechanism (one-to-one, when redundancy is not met, many-to-one), and transfers part of the charging demand from the overloaded area to the adjacent area to achieve load balancing and resource optimization. The load assessment module not only realizes real-time early warning of overload conditions, but also provides intelligent scheduling and resource sharing functions, effectively improving the operation safety and service quality of the charging network, and preventing system failures or user experience degradation due to local overload.
[0066] The present invention also provides a charging demand prediction method based on multi-source heterogeneous data fusion, which includes the following steps:
[0067] Step 1: Collect charging-related data in the current area and pre-process the charging-related data to obtain processed charging-related data;
[0068] Step 2: Input the processed charging-related data into the deep learning unit to obtain a preliminary prediction value
[0069] Step 3: Analyze the historical charging demand in the current area and obtain the historical charging impact coefficient;
[0070] Step 4: Analyze the convenience of the current area and obtain the impact coefficient of charging convenience;
[0071] Step 5: Use the electricity demand forecast module to read the historical electricity demand database to obtain the current area's historical charging demand;
[0072] Step 6: Obtain the final charging forecast value based on a comprehensive analysis of the preliminary forecast value, the historical charging impact coefficient, the charging convenience impact coefficient, and the current area's historical charging demand;
[0073] Step 7: Evaluate the final charging prediction value, issue an early warning based on the charging prediction value, and implement the early warning strategy.
[0074] 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. A charging demand forecasting system based on multi-source heterogeneous data fusion, characterized by: It includes data acquisition module, preliminary prediction module, charging demand prediction module and load assessment module; The data acquisition module is used to collect charging-related data in the current area and pre-process the charging-related data to obtain processed charging-related data; The preliminary prediction module is equipped with a deep learning unit, and the processed charging-related data is input into the deep learning unit to obtain a preliminary prediction value CBYCZ; The charging demand prediction module includes a historical charging demand analysis unit, a charging convenience evaluation unit, and a charging demand prediction unit. The historical charging demand analysis unit analyzes the historical charging demand of the current area to obtain a historical charging impact coefficient. The charging convenience evaluation unit is used to analyze the convenience of the current area to obtain a charging convenience impact coefficient. The charging demand prediction unit obtains a final charging prediction value ZZYCZ based on a comprehensive analysis of the preliminary prediction value, the historical charging impact coefficient, the charging convenience impact coefficient, and the historical charging demand of the current area. The load evaluation module is used to evaluate the final charging prediction value, issue an early warning based on the charging prediction value, and execute an early warning strategy.
2. The multi-source heterogeneous data fusion charging demand forecasting system according to claim 1 is characterized by: The data acquisition module collects charging-related data of the current area, including temperature, humidity, precipitation probability, user charging time, SOC during charging, and charging power of the charging pile.
3. The multi-source heterogeneous data fusion charging demand forecasting system according to claim 2 is characterized by: The charging-related data is preprocessed using Z-score standardization, and the deviation of the preprocessed values from their mean is analyzed, and values beyond k standard deviations from the mean after preprocessing are screened out to obtain the processed charging-related data.
4. The multi-source heterogeneous data fusion charging demand forecasting system according to claim 1 is characterized by: The deep learning unit is specifically a graph attention network, and the input is processed charging-related data, the output is power, and the output power is used as a preliminary prediction value.
5. The multi-source heterogeneous data fusion charging demand forecasting system according to claim 1 is characterized by: The historical charging demand analysis unit analyzes the historical charging demand of the current area and obtains the specific expression of the historical charging impact coefficient as follows: Among them, LSCDYX represents the historical charging impact coefficient of the i-th evaluation, XQZ n Indicates the nth environment consistency coefficient. If the current area is consistent with the nth environment in the statistics of the actual charging demand for the i-th time, then XQZ n =1, otherwise XQZ n =0.
6. The multi-source heterogeneous data fusion charging demand forecasting system according to claim 5 is characterized by: The charging convenience evaluation unit is used to analyze the convenience of the current area, and obtain the specific expression of the charging convenience influence coefficient as follows: Among them, PCZ h It represents the h-th deviation value that affects the user's arrival at the current area. The higher the value, the worse the convenience. CDBJ represents the charging convenience impact coefficient.
7. The multi-source heterogeneous data fusion charging demand forecasting system according to claim 6 is characterized by: The current area historical charging demand is obtained by reading the historical power demand database through the charging demand prediction module, and the specific steps are as follows: Read the historical power demand database based on LSCDYX i Filter one by one and select all LSCDYX i The top M actual charging demands in the ranking are averaged as the historical charging demand of the current region 8. The multi-source heterogeneous data fusion charging demand forecasting system according to claim 7 is characterized by: The charging demand forecasting unit obtains the final charging forecast value ZZYCZ based on a comprehensive analysis of the preliminary forecast value, the historical charging impact coefficient, the charging convenience impact coefficient, and the current area's historical charging demand. The specific expression is as follows: in, represents the historical charging demand in the current area, CBYCZ represents the preliminary forecast value, ZZYCZ represents the final charging forecast value, and CDBJ represents the charging convenience impact coefficient.
9. The charging demand forecasting system based on multi-source heterogeneous data fusion according to claim 1 is characterized by: The load assessment module is used to assess the charging prediction value, issue an early warning based on the charging prediction value, and implement the early warning strategy in the following specific steps: When the final charging prediction value ZZYCZ exceeds the preset charging load, an early warning is issued, and the redundancy of adjacent charging areas is obtained, and scheduling is carried out by the adjacent areas.
10. A method for predicting charging demand based on multi-source heterogeneous data fusion, based on the system for predicting charging demand based on multi-source heterogeneous data fusion according to any one of claims 1 to 9, characterized in that: The following steps are involved: Step 1: Collect charging-related data in the current area and pre-process the charging-related data to obtain processed charging-related data; Step 2: Input the processed charging-related data into the deep learning unit to obtain a preliminary prediction value Step 3: Analyze the historical charging demand in the current area and obtain the historical charging impact coefficient; Step 4: Analyze the convenience of the current area and obtain the impact coefficient of charging convenience; Step 5: Read the historical power demand database to obtain the current area's historical charging demand; Step 6: Obtain the final charging forecast value based on a comprehensive analysis of the preliminary forecast value, the historical charging impact coefficient, the charging convenience impact coefficient, and the current area's historical charging demand; Step 7: Evaluate the final charging prediction value, issue an early warning based on the charging prediction value, and implement the early warning strategy.