New energy charging pile intelligent power control method and system based on artificial intelligence

Abstract

The invention provides a new energy charging pile intelligent power control method and system based on artificial intelligence, and the method comprises the steps: obtaining user side data, environment context and pile position state information, and determining a corresponding actual expected power load; embedding the actual expected power load into a graph structure to carry out heat potential diffusion analysis to obtain a heat potential value of a corresponding node; a resource scheduling optimization function is designed based on heat potential diffusion analysis, and the resource scheduling optimization function determines a power distribution value which can be actually obtained by each pile position according to the heat potential value and the constraint condition; and issuing the actually available power distribution value of each pile position to actual charging pile control equipment, driving the actual charging pile control equipment to execute a power output behavior according to a system strategy in a current scheduling period, monitoring the actual control execution condition of each pile position in real time, and performing dynamic adjustment when necessary.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and particularly relates to an intelligent power control method and system for new energy charging piles based on artificial intelligence. Background Art

[0002] With the large-scale popularization of new energy electric vehicles, charging piles, as key infrastructure, have gradually become an important part of the urban energy network. Especially in scenarios such as urban centralized fast charging stations, logistics hubs, and residential communities, multiple charging pile systems are connected in parallel to the local power grid, significantly increasing the complexity of power scheduling and control. At the same time, the high proportion of new energy power (such as wind power and photovoltaic power) access has also changed the stability basis of the power system. Its power generation output has significant volatility and unpredictability, resulting in the frequent failure of traditional control strategies based on stable power supply in actual deployment. Furthermore, the current user usage behavior shows highly non-linear and uncertain characteristics. For example, there are differences between the reserved charging time and the actual arrival time of users, mismatches between the reserved power and the actual power demand, and frequent occurrences of phenomena such as some users defaulting on appointments or temporarily inserting charging piles. Existing charging systems generally adopt fixed strategies or simplified rules for power pre-allocation, such as guiding peak-valley electricity prices based on time periods and locking power according to reserved time. These methods can play a certain role under the assumptions of ideal user behavior and stable energy supply, but it is difficult to adapt to the rapidly changing load demand in a real dynamic environment, thereby resulting in waste of electric energy resources, local overload of the power grid, and even scheduling failures.

[0003] In addition, since charging piles are usually arranged in a grid-like or locally aggregated layout in terms of physical structure, there is a certain power coupling effect in the circuit connection between different pile positions, which means that the power adjustment of a certain pile position may have a chain effect on adjacent pile positions. However, most of the existing power control systems are based on the ideas of "independent for each pile" or "global unified optimization", and fail to effectively capture and utilize these local structural relationships. Therefore, the system often has problems of unbalanced resource allocation in local high-load scenarios: some pile positions have power overflow and power is limited, while some pile positions are in an idle state; and this imbalance cannot be avoided through simple weighting or linear optimization strategies. In addition, due to the dynamic evolution characteristics of electric vehicle charging behavior in time and space, the system needs to have a highly flexible and real-time updated control ability. However, the "prediction-control" decoupled two-stage process adopted by most existing systems often loses the adjustment space when user behavior fails to be fulfilled on schedule or the input on the power grid side changes, and it is difficult to meet the current high requirements for intelligent scheduling and refined management.

[0004] Therefore, the existing charging control methods still have obvious defects in dealing with new energy fluctuations, user behavior uncertainty, and multi-pile system collaborative scheduling. There is an urgent need for a new control system with higher real-time performance, self-adaptability, and intelligent level. Summary of the Invention

[0005] The object of the present invention is to propose an intelligent power control method and system for new energy charging piles based on artificial intelligence, which can effectively solve the prominent defects of existing methods in aspects such as insufficient dynamic perception of activity, low intelligence of sub-packaging scheduling, and lack of ability to cope with complex sub-packaging environments.

[0006] To achieve the above object, in the first aspect of the present invention, there is provided an intelligent power control method for new energy charging piles based on artificial intelligence, and the method includes the following steps:

[0007] Obtain user-side data, environmental context, and pile position status information, combine them into a feature vector of the current pile position, input the feature vector into a neural network, output the probability that a user actually appears and starts charging within the current scheduling period of the corresponding pile position and the predicted true power load value, and determine the corresponding actual expected power load. The actual expected power load is the weighted product of the probability that a user actually appears and starts charging within the current scheduling period of the corresponding pile position and the predicted true power load value;

[0008] Embed the actual expected power load into a graph structure for heat potential diffusion analysis to obtain the heat potential value of the corresponding node;

[0009] Design a resource scheduling optimization function based on heat potential diffusion analysis. The resource scheduling optimization function determines the actually available power allocation value for each pile position according to the heat potential value and constraint conditions; the constraint conditions include that the total power scheduling of the whole station shall not exceed the system available value, the allocated power of each pile position shall not exceed its access physical limit, and the system no longer allocates power to pile positions with extremely low confidence to avoid resource waste caused by behavior prediction errors;

[0010] Send the actually available power allocation value for each pile position to the actual charging pile control device, drive it to execute the power output behavior according to the system strategy within the current scheduling period, and monitor the actual execution situation of the control of each pile position in real time, and make dynamic adjustments when necessary.

[0011] Preferably, the user-side data includes the user's reservation history, current reservation information, and user ID or vehicle type label; the environmental context includes weather, holiday flag, and whether power is restricted in this area; the pile position status information includes the current operating load of the pile and whether the connected power source is a new energy source.

[0012] Preferably, the neural network is a trainable neural network model. Among them, a behavior deviation penalty term is introduced during the training of the neural network to punish the negative impact of users with a high no-show rate on system scheduling; the loss function of the neural network is determined based on binary cross-entropy loss and the behavior deviation penalty term.

[0013] Preferably, the graph structure is constructed as follows:

[0014] The node set represents each charging station in the station, and the attributes of the node include the actual expected power load;

[0015] The edge set represents the connection relationship between nodes, and the edge weight is used to describe the physical coupling strength between adjacent nodes. Its data sources include power topology drawings, GIS spatial distance matrices, or actual resistance coefficients of site wiring.

[0016] Preferably, the thermal potential diffusion analysis is expressed as:

[0017]

[0018] Among them, H i Represents node v i The thermal potential value represents the system's overall assessment of the power pressure of the pile position in the current cycle; φ i is the actual expected power load of the node, indicating its own power heat source intensity; w ij v i Its adjacent node v j The coupling strength between k w ik is the normalization coefficient, which prevents high-connectivity nodes from absorbing excessive energy; is the set of all adjacent nodes of node i; δ i is the maximum new energy fluctuation value detected in the branch connected to the pile during this period; η i It is a system-set fluctuation interference coefficient used to adjust the response sensitivity of new energy access to thermal potential disturbances.

[0019] Preferably, a thermal gradient penalty mechanism is introduced into the graph structure to adjust the extremely unbalanced thermal potential distribution.

[0020] Preferably, the source scheduling optimization function is expressed as:

[0021]

[0022] in, is the final power distribution value of pile position i; H i is the thermal potential of the pile; ∈ is a very small constant to prevent division by zero; Ω i It is the sum of the squares of the local thermal potential gradients between the pile and the adjacent nodes, and is used to indicate whether the thermal state of the point is out of balance with the surroundings. A large value indicates structural imbalance in the region. Represents the historical average charging power of the user of the pile, which is used to define the expected energy supply range of the system; φ iis the actual expected power load; β and λ are the structural balance control weight and the behavior deviation penalty weight respectively;

[0023] Among them, represents the driving mechanism for the scheduling resource to reversely regulate the thermal potential. The nodes with higher thermal potential get less power; β·Ω i is used to suppress the area with too high local gradient; is the behavior deviation penalty term, which is used to punish the scheduling decision with too large deviation between the currently allocated power and the historical power when the confidence level is relatively high.

[0024] Preferably, the step of distributing the actually available power allocation value of each pile position to the actual charging pile control device includes:

[0025] Packaging the following parameters into a control instruction through a standard protocol and sending it to each pile control unit: the target power of pile position i in the current cycle, the length of the scheduling cycle, the maximum allowable dynamic fluctuation to prevent power mutation from damaging the equipment, and the expected user plug-in delay time.

[0026] Preferably, at the end of the scheduling cycle, the actual average output power of each pile is transmitted back through the power acquisition module:

[0027] If the actual average output power of each pile position exceeds the system tolerance threshold, it means that the pile fails to execute the power task as planned, and the system triggers a fine-tuning mechanism; the triggering of the fine-tuning mechanism is as follows: the maximum allowable dynamic fluctuation is regarded as recoverable power resources, and the system will reallocate this part of the power within the current cycle, preferentially supplying the pile positions with high actual expected power load and low thermal potential.

[0028] In the second aspect of the present invention, there is provided an intelligent power control system for new energy charging piles based on artificial intelligence, and the system includes:

[0029] A user behavior credibility modeling module, which is used to obtain user-side data, environmental context and pile position status information, combine them into a feature vector of the current pile position, input the feature vector into a neural network, output the probability that the user actually appears and starts charging within the current scheduling cycle of the corresponding pile position and the predicted true power load value, and determine the corresponding actual expected power load, where the actual expected power load is the weighted product of the probability that the user actually appears and starts charging within the current scheduling cycle of the corresponding pile position and the predicted true power load value;

[0030] A spatial thermal potential map construction module, which is used to embed the actual expected power load into a graph structure for thermal potential diffusion analysis to obtain the thermal potential value of the corresponding node;

[0031] A resource scheduling optimization module, which is used to design a resource scheduling optimization function based on heat potential diffusion analysis. The source scheduling optimization function determines the actual power allocation value that each pile position can obtain according to the heat potential value and constraint conditions. The constraint conditions include that the total power scheduling of the whole station cannot exceed the system available value, the allocated power of each pile position cannot exceed its access physical limit, and the system no longer allocates power to pile positions with extremely low confidence, so as to avoid resource waste caused by behavior prediction errors.

[0032] A control instruction execution and feedback module, which is used to send the actual power allocation value that each pile position can obtain to the actual charging pile control device, drive it to execute the power output behavior according to the system strategy within the current scheduling period, and monitor the actual execution situation of the control of each pile position in real time, and make dynamic adjustments when necessary.

[0033] The beneficial technical effects of the present invention are at least as follows:

[0034] Aiming at the problems existing in the current charging pile power control system in aspects such as new energy fluctuations, user behavior uncertainty, and inter-pile coupling scheduling, the present invention proposes an intelligent power control method and system with adaptive adjustment ability for the background of new energy power supply. Based on multi-source perception data, the invention can dynamically evaluate the credibility of user behavior, and then accurately identify the real load demand of the system in the future period of time, realizing the predictive guidance of power allocation. At the same time, the system can model the spatial structure and power coupling relationship between charging piles, so that power scheduling not only performs global optimization based on the total demand and grid constraints, but also can dynamically respond to local power pressure, realizing the coordinated control of heat zone load reduction and cold zone power guidance. By constructing a unified control framework, the present invention breaks through the barrier between user behavior prediction and physical scheduling control, forms a prediction-control integrated intelligent scheduling mechanism, and effectively improves the control stability and power utilization rate under the conditions of highly uncertain power resource fluctuations and usage behaviors. The system has real-time response ability, strong structural adaptability, and can adapt to multi-pile networks with different layouts and scales, and is especially suitable for large-scale charging pile scenarios under the background of high proportion of new energy grid connection, which is significantly better than the existing traditional control methods. Description of the Drawings

[0035] The present invention is further described with the accompanying drawings, but the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the following drawings without creative efforts.

[0036] Figure 1 It is a flowchart of the intelligent power control method for new energy charging piles based on artificial intelligence of the present invention. Detailed Embodiments

[0037] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0038] As Figure 1 shown, the artificial intelligence-based new energy charging pile intelligent power control method provided by the embodiment of the present invention includes:

[0039] S1. Obtain user-side data, environmental context, and pile position status information, combine them into a feature vector of the current pile position, input the feature vector into a neural network, output the probability that a user actually appears and starts charging within the current scheduling period of the corresponding pile position and the predicted true power load value, and determine the corresponding actual expected power load. The actual expected power load is the weighted product of the probability that a user actually appears and starts charging within the current scheduling period of the corresponding pile position and the predicted true power load value.

[0040] Specifically, in the new energy charging pile system, due to the uncertainty fluctuations on the power supply side (such as changes in sunlight for photovoltaic power supply) and the behavioral uncertainties on the user side (appointment non-compliance, temporary plugging in of piles, low usage rate, etc.), if the control system still schedules according to "the time period and power reserved by each user", it will inevitably cause waste of resources, power overflow, or concentration of peak pressure, ultimately leading to scheduling failure. Therefore, the goal of this step is to construct a dynamic and reliable load estimation method, accurately model "whether the user will actually use this pile position" and "what the power load will be when actually used", and use this value as the "heat source input" in the entire control system to drive the subsequent structured scheduling strategy.

[0041] Among them, the input of this step consists of three aspects: First is the user-side data, including the user's reservation history (such as on-time arrival rate, average actual usage power, number of no-shows, etc.), current reservation information (reservation time, duration, power), and user ID or vehicle type label; second is the environmental context, such as weather (heavy rainfall or extreme high temperature will affect travel), holiday flag, whether power is restricted in this area and other macroscopic variables; finally is the pile position status information, such as the current operating load of this pile, whether the connected power source is a new energy source, etc. These data are combined into the feature vector xi of pile position i i , which is processed by the neural network f θ , and outputs two core variables: representing the probability that a user actually appears and starts charging at this pile within the current scheduling period T, P i representing the predicted true power load value (unit: kW). Among them, f θIt is a trainable neural network model, and its structure should be able to support heterogeneous inputs (for example, using the attention mechanism to process time information, or using embeddings to represent user classification features).

[0042] When calculating the actual expected power load φ i Instead of simply We introduce a dynamic adjustment term γ for the new energy background in this estimate i to improve the adaptability to power supply pressure:

[0043]

[0044] Where:

[0045] φ i : The confidence power heat source value of the i-th pile position, representing the credible power demand of this pile in the current cycle;

[0046] The model f θ The probability that the user actually appears and charges, predicted based on the x i Feature prediction;

[0047] P i : The estimated actual power load of this user calculated based on historical behavior (unit: kW);

[0048] γ i : The new energy adjustment factor set by the system, representing the sensitivity of the system to new energy fluctuations (empirical setting or optimized through cross-validation);

[0049] R i : The new energy power supply volatility of this pile in the current cycle (such as the photovoltaic power change rate);

[0050] The average power supply volatility of all piles in the entire area, used for normalization processing.

[0051] To improve the credible generalization ability of the model, we introduce a behavior deviation penalty term when training f θ to punish the negative impact on system scheduling caused by users with a high no-show rate. The specific loss function is:

[0052]

[0053] Where:

[0054] The overall loss function;

[0055] L bce : The standard binary cross-entropy loss;

[0056] yi : The tag value indicating whether the user has actually arrived;

[0057] α i : The behavioral instability coefficient of the i-th user, such as the proportion of no-show times in its recent n reservations (0 - 1);

[0058] λ: The regularization term weight, which controls the intensity of the deviation penalty term.

[0059] The core variable φ output in this step i represents the "heat source intensity" of this pile position within this scheduling period, and is directly input into the next spatial scheduling diagram as a node attribute, representing the true power load that the system is expected to bear at this node during the current period. This quantity not only reflects the prediction of user behavior but also integrates the real-time impact of new energy power supply status on the scheduling strategy, thereby forming the initial driving term for the closed-loop coordination of prediction - status - control. Different from the traditional approach of treating user behavior prediction as an "external module", this solution deeply integrates the behavior credibility with the power scheduling system, making the scheduling status perceived by the system more physically meaningful and dynamically adjustable.

[0060] S2. Embed the actual expected power load into the graph structure for heat potential diffusion analysis to obtain the heat potential values of the corresponding nodes.

[0061] Specifically, the role of this step is to embed the confidence power value φ of each pile position output in the previous step i (i.e., the weighted product of the probability that the user actually uses this pile and the power demand within the current scheduling period) into the spatial network structure of the entire charging station to construct a spatial heat potential map that can reflect "local power pressure", "multi-pile power coupling", and "new energy access volatility". Different from the traditional centralized distribution strategy that only targets the total power demand, this step constructs a graph structure G=(V, E) and executes a physically inspired heat potential diffusion mechanism on it to achieve spatial modeling and structural reasoning of the energy demand of the entire station, providing an energy distribution view that is highly close to the physical structure of the station for subsequent power optimization scheduling. This step is the key bridge to establish a logical connection between "user behavior prediction" and "control execution" and serves as the middle-state modeler of the system.

[0062] Furthermore, first construct the graph structure G=(V, E). Among them, the node set V represents each charging pile position v within the station i , and the attributes of the nodes include the actual expected power load φ i , and its source is the value output according to the user behavior prediction model f in step one θ after being adjusted by new energy fluctuations. The edge set E represents the connection relationship between nodes, and the edge weight w is used to describe v ij ​i The physical coupling strength with v j where the data sources include power topology drawings, GIS spatial distance matrices, or the actual resistance coefficients of substation wiring, etc.

[0063] To model the impact of new - energy power volatility on the local power distribution capacity, we map φ i as a "heat source" and inject it into each node, and construct H based on the following heat - potential diffusion formula i , that is, the energy - load pressure perception value of each node in the heat - potential map:

[0064]

[0065] Where:

[0066] H i represents the heat - potential value of node v i , which is the output of this step and represents the overall assessment of the system's current - cycle power pressure on this pile position;

[0067] φ i is the actual expected power load of this node v i , representing its own power heat - source intensity, sourced from step one;

[0068] w ij is the coupling strength between node v i and its adjacent node v j , usually normalized to the reciprocal of 1 meter or calculated from the wiring resistance;

[0069] ∑ k w ik is the normalization coefficient to prevent high - connectivity nodes from over - absorbing energy;

[0070] is the set of all adjacent nodes of node v i ;

[0071] δ i is the maximum new - energy fluctuation value detected in this cycle in the branch where this pile is connected (e.g., the instantaneous standard deviation of the current on the access side in the photovoltaic access area);

[0072] η i is a system - set "fluctuation interference coefficient" that adjusts the response sensitivity of new - energy access to heat - potential perturbations.

[0073] This formula has three innovations compared with traditional graph - propagation methods:

[0074] Firstly, introducing φ i as a "structural heat source" enables the credibility of user behavior to directly drive the heat - field propagation;

[0075] Second, through the neighbor confidence φ j The weighted propagation simulates the local power accumulation area and can identify local high-risk power aggregation points;

[0076] Third, innovatively add the -δ i ·η i term to achieve "thermal potential suppression" of the new energy fluctuation area, reflecting the scheduling conservatism of the system in severely fluctuating areas, and effectively preventing the formation of pseudo hot spots in the fluctuation area.

[0077] To enhance the system's recognition ability for overheated areas (i.e., areas with abnormal power aggregation), we further construct a local hot spot regularization term as a correction mechanism for the thermal potential of the graph structure. Specifically, the following thermal gradient penalty mechanism is introduced into the graph to adjust the extremely unbalanced thermal potential distribution:

[0078]

[0079] where: H j is the thermal potential value of node v j ;

[0080] Ω i is the local thermal potential regularization term for node v i used to characterize the sum of squares of the differences in thermal potential between it and the surrounding nodes;

[0081] μ is a regularization weight coefficient (such as 0.1 - 1.0), controlling the strength of the thermal potential equilibrium trend.

[0082] This term does not directly act on the numerical calculation of H i , but will be used as a penalty factor in subsequent scheduling optimization for policy selection. Its practical significance is: encouraging the scheduling system to preferentially allocate loads to areas with low thermal potential and gentle thermal potential gradients when allocating power in the future, improving scheduling security and system energy balance.

[0083] The final output of this step is the thermal potential value H i of each node, accompanied by the local gradient regularization value Ω i . H i is used as an evaluation of the power load pressure of each pile under the current scheduling cycle and is the target input value of the subsequent optimization module, while Ω i provides an "energy adjustment strategy preference in space" for the optimizer.

[0084] S3. Design a resource scheduling optimization function based on heat potential diffusion analysis. The source scheduling optimization function determines the actual power allocation value that each pile position can obtain according to the heat potential value and constraint conditions. The constraint conditions include that the total power scheduling of the whole station shall not exceed the system available value, the allocated power of each pile position shall not exceed its access physical limit, and the system no longer allocates power to the pile positions with extremely low confidence, so as to avoid wasting resources due to behavior prediction errors.

[0085] Specifically, the key role of this step in the entire patent system is to transform the result of the previous step of "modeling based on the heat potential state of structure and user behavior" into a "power scheduling strategy with control and execution significance". Specifically, the heat potential value H of each charging pile has been output in the previous stage i , which reflects the "pressure" state of energy use under the combined action of the user confidence demand φ i of each pile position in the current cycle, the spatial topological coupling w ij , and the new energy fluctuation δ i . Now the task of this step is to determine the actual power allocation value that each pile position can obtain on the basis of considering factors such as power supply capacity, grid load safety, and user credibility Traditional strategies mostly adopt methods such as "average distribution", "filling with maximum power", or "linear mapping of reserved power", which cannot respond to the heat potential state and cannot constrain the negative impacts brought by new energy volatility. Therefore, this step introduces multiple innovative scheduling mechanisms for scenarios to make the output have behavior guidance, structural adaptability, stability and physical feasibility.

[0086] Among them, first clarify the optimization goal. We hope to allocate the current available total power P of the system total among all pile positions and achieve two major goals on the premise of meeting the following constraints: (1) Reduce the total pressure of power heat potential of the whole station, and preferentially allocate to the pile positions with low heat potential and high confidence, so as to avoid hot spot aggregation and resource redundancy; (2) Keep the spatial distribution balanced and avoid the phenomenon of "excessive local pressure" in some areas. Based on this, define the following optimization objective function:

[0087]

[0088] Where:

[0089] is the final power allocation value of pile position i and is the output of this step;

[0090] H i is the heat potential value of this pile, which is obtained from the previous step of spatial heat field reasoning. The higher the value, the less the system currently hopes that this area will continue to obtain energy;

[0091] ∈ is a minimum constant to prevent division by zero;

[0092] Ω i is the sum of the squares of the local thermal potential gradients between this pile and adjacent nodes (calculated in the previous step), used to represent "whether the heat state at this point is out of balance with the surroundings". A large value indicates structural imbalance in this area;

[0093] represents the average historical charging power of this pile's users, used to define the expected energy supply range of the system;

[0094] β and λ are the structural balance control weight and the behavior deviation penalty weight respectively.

[0095] It can be understood that the innovation points of this optimization goal include:

[0096] The first term represents the driving mechanism for the scheduling resource to reversely adjust the thermal potential. Nodes with higher thermal potential receive less power, thus forming a "diversion" mechanism to control local energy aggregation;

[0097] The second term β·Ω i is used to suppress areas with too high local gradients, forming a "structural balance orientation" to ensure the controllability of energy flow in space in the system;

[0098] The third term is the behavior deviation penalty term, used to penalize scheduling decisions with too large a deviation between the currently allocated power and the historical power when the confidence level is relatively high. This can effectively prevent the "over-regulation" in the high-confidence area from degrading the user experience, and at the same time automatically reduce the penalty intensity when the confidence level is low, reflecting the system's response ability to the confidence level of behavior prediction.

[0099] Based on the above optimization goal, the following feasibility constraints need to be added:

[0100] The total power scheduling of the entire substation shall not exceed the available value of the system;

[0101] The allocated power of each pile position cannot exceed its access physical limit;

[0102] When φ i <φ min (set to 0.05 for example), force to set That is, the system no longer allocates power to pile positions with extremely low confidence levels, avoiding waste of resources caused by behavior prediction errors.

[0103] This scheduling optimization problem is a convex problem with boundary constraints and can be solved by the conventional projected gradient descent method (PGD). The system measures the current thermal potential state H every fixed scheduling period (such as 5 minutes) iWith confidence φ i Update and recalculate the objective function to achieve rolling scheduling. To enhance the system's adaptability to sudden new energy fluctuations, the system allows dynamic adjustment of β and λ during extreme new energy disturbances (such as ) to enable the system to enter the "suppress scheduling fluctuations" safety mode. The final output of this step is the power allocation value of all pile positions This value will be directly passed to the control execution module for generating physical power control signals.

[0104] S4. Distribute the power allocation value actually available for each pile position to the actual charging pile control device, drive it to perform power output behavior according to the system strategy within the current scheduling period, and monitor the actual execution status of each pile position in real time, and make dynamic adjustments when necessary.

[0105] Specifically, the goal of this step is to distribute the power allocation value output by the intelligent optimization scheduling module in the previous stage to the actual charging pile control device, drive it to perform power output behavior according to the system strategy within the current scheduling period T (such as 5 minutes). At the same time, the system needs to introduce an execution feedback mechanism, monitor the actual execution status of each pile position in real time, and make dynamic adjustments when necessary, forming a complete closed-loop structure of "scheduling → execution → feedback → fine-tuning" to ensure that the intelligent control system has physical deployability, abnormal self-repair ability, and scheduling stability.

[0106] Among them, the input of control execution is the scheduling power output in step three (unit: kW). The system packs the following parameters into control instructions through standard protocols (such as Modbus / TCP, CAN, IEC 61850) and sends them to each pile control unit:

[0107] T: Scheduling period length (such as 5 minutes);

[0108] Δ i : Maximum allowable dynamic fluctuation to prevent power mutation from damaging equipment;

[0109] d i : Expected user plug-in pile delay time (if there is a prediction of the user arrival time at this pile in step one);

[0110] The controller realizes precise adjustment of the power output of the pile position through methods such as converter adjustment and power segment control.

[0111] At the end of the scheduling period, the pile control system returns the actual average output power of each pile through the power acquisition module The system calculates the current scheduling execution deviation:

[0112]

[0113] Wherein:

[0114] e i is the execution deviation rate of pile position i;

[0115] is the true average output collected;

[0116] ∈ is a minimum constant (to prevent division by zero), such as 10 -4 .

[0117] If |e i | exceeds the system tolerance threshold η (such as 10%), it indicates that the pile does not execute the power task as planned, and the system triggers a fine-tuning mechanism. At this time, the unexecuted part is regarded as "recoverable power resources". The system will reallocate this part of the power within the current cycle, and preferentially supply pile position j with high confidence (high actual expected power load φ j high) and low heat potential (H j low). The allocation weight is defined as follows:

[0118]

[0119] This mechanism ensures that the resource replenishment has a dual priority logic of "structural pressure guidance + user behavior reliability", avoiding excessive disturbances to the system as a whole caused by scheduling resource waste or single-point failure.

[0120] All execution states (power response, deviation, reasons for unused power, whether the pile is inserted, etc.) are recorded in the operation log of the scheduling system for subsequent scheduling visualization, system maintenance analysis, and abnormal alarm.

[0121] The embodiment of the present invention also provides an intelligent power control system for new energy charging piles based on artificial intelligence. The system includes:

[0122] A user behavior credibility modeling module, which is used to obtain user-side data, environmental context, and pile position status information, combine them into a feature vector of the current pile position, input the feature vector into a neural network, output the probability that the user actually appears and starts charging within the current scheduling cycle of the corresponding pile position and the predicted true power load value, and determine the corresponding actual expected power load. The actual expected power load is the weighted product of the probability that the user actually appears and starts charging within the current scheduling cycle of the corresponding pile position and the predicted true power load value;

[0123] A spatial heat potential map construction module, which is used to embed the actual expected power load into a graph structure for heat potential diffusion analysis to obtain the heat potential value of the corresponding node;

[0124] A resource scheduling optimization module is used to design a resource scheduling optimization function based on heat potential diffusion analysis. The source scheduling optimization function determines the actual available power allocation value for each pile position according to the heat potential value and constraint conditions. The constraint conditions include that the total power scheduling of the whole station shall not exceed the system available value, the allocated power of each pile position shall not exceed its access physical limit, and the system no longer allocates power to pile positions with extremely low confidence to avoid resource waste caused by behavior prediction errors.

[0125] A control instruction execution and feedback module is used to send the actual available power allocation value of each pile position to the actual charging pile control device, drive it to execute the power output behavior according to the system policy within the current scheduling period, and monitor the actual execution situation of each pile position in real time, and make dynamic adjustments when necessary.

[0126] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0127] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces, and the indirect coupling or communication connection of the device or unit can be in an electrical, mechanical, or other form.

[0128] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical disks, etc., which can store program codes.

[0129] Although embodiments of the present invention have been shown and described, those skilled in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the claims and their equivalents.

Claims

1. An intelligent power control method for new energy charging piles based on artificial intelligence, characterized in that: The method comprises the following steps: Obtain user-side data, environmental context, and pile position status information, combine them into a feature vector of the current pile position, input the feature vector into a neural network, output the probability of the user actually appearing and starting charging and the estimated real power load value of the corresponding pile position in the current scheduling period, and determine the corresponding actual expected power load, which is the weighted product of the probability of the user actually appearing and starting charging and the estimated real power load value in the current scheduling period of the corresponding pile position; The actual expected power load is embedded in the graph structure to perform thermal potential diffusion analysis to obtain the thermal potential value of the corresponding node; A resource scheduling optimization function is designed based on thermal potential diffusion analysis. The resource scheduling optimization function determines the actual power allocation value available for each pile position according to the thermal potential value and constraint conditions. The constraint conditions include that the total power scheduling of the entire station shall not exceed the system available value, the allocated power of each pile position shall not exceed its access physical limit, and the system will no longer allocate power to pile positions with extremely low confidence, so as to avoid waste of resources due to behavior prediction errors. The power allocation value actually available for each charging pile position is sent to the actual charging pile control device, driving it to execute power output behavior according to the system strategy during the current scheduling cycle, and monitoring the actual execution of the control of each charging pile position in real time, and making dynamic adjustments based on the actual execution situation.

2. The intelligent power control method for new energy charging piles based on artificial intelligence according to claim 1 is characterized in that: The user-side data includes the user's reservation history, current reservation information, and user ID or vehicle model label; the environmental context includes weather, holiday signs, and whether there is power restriction in the current area; the pile position status information includes the current operating load of the pile and whether the connected power source is a new energy source.

3. The intelligent power control method for new energy charging piles based on artificial intelligence according to claim 1 is characterized in that: The neural network is a trainable neural network model, wherein a behavior deviation penalty term is introduced during the training of the neural network to punish the negative impact of users with high no-show rates on system scheduling; the loss function of the neural network is determined based on the binary cross entropy loss and the behavior deviation penalty term.

4. The intelligent power control method for new energy charging piles based on artificial intelligence according to claim 1 is characterized in that: The graph structure is constructed as follows: The node set represents each charging station in the station, and the attributes of the node include the actual expected power load; The edge set represents the connection relationship between nodes, and the edge weight is used to describe the physical coupling strength between adjacent nodes. Its data sources include power topology drawings, GIS spatial distance matrices, or actual resistance coefficients of site wiring.

5. The intelligent power control method for new energy charging piles based on artificial intelligence according to claim 4 is characterized in that: The thermal potential diffusion analysis is expressed as: Among them, H i Represents node v i The thermal potential value represents the system's overall assessment of the power pressure of the pile position in the current cycle; φ i is the actual expected power load of the node, indicating its own power heat source intensity; w ij v i Its adjacent node v j The coupling strength between k w ik is the normalization coefficient, which prevents high-connectivity nodes from absorbing excessive energy; is the set of all adjacent nodes of node i; δ i is the maximum new energy fluctuation value detected in the branch connected to the pile during this period; η i It is a system-set fluctuation interference coefficient used to adjust the response sensitivity of new energy access to thermal potential disturbances.

6. The intelligent power control method for new energy charging piles based on artificial intelligence according to claim 5 is characterized in that: A thermal gradient penalty mechanism is introduced into the graph structure to adjust the extremely unbalanced thermal potential distribution.

7. The intelligent power control method for new energy charging piles based on artificial intelligence according to claim 1 is characterized in that: The source scheduling optimization function is expressed as: in, is the final power distribution value of pile position i; H i is the thermal potential of the pile; ∈ is a very small constant to prevent division by zero; Ω i It is the sum of the squares of the local thermal potential gradients between the pile and the adjacent nodes, and is used to indicate whether the thermal state of the point is out of balance with the surroundings. A large value indicates structural imbalance in the current region. Represents the historical average charging power of the user of the pile, which is used to define the expected energy supply range of the system; φ i is the actual expected power load; β and λ are the structural equilibrium control weight and the behavior deviation penalty weight respectively; in, It represents the driving mechanism of scheduling resources to reversely regulate the thermal potential. The node with the highest thermal potential gets the least power. i Used to suppress areas where local gradients are too high; It is a behavior deviation penalty term, which is used to penalize the scheduling decision when the current allocated power deviates too much from the historical power when the confidence is highest.

8. The intelligent power control method for new energy charging piles based on artificial intelligence according to claim 1 is characterized in that: The step of sending the power allocation value actually available at each charging pile position to the actual charging pile control device includes: The following parameters are packaged as control instructions through the standard protocol and sent to each pile control unit: the target power of the current cycle at pile position i, the length of the scheduling cycle, the maximum allowed dynamic fluctuation to prevent power mutation from damaging the equipment, and the expected user plug-in delay time.

9. The intelligent power control method for new energy charging piles based on artificial intelligence according to claim 8 is characterized in that: At the end of the scheduling cycle, the actual average output power of each pile is transmitted back through the power collection module: If the actual average output power of each pile position exceeds the system tolerance threshold, it means that the current pile position has not performed the power task as planned, and the system triggers the fine-tuning mechanism; the triggering fine-tuning mechanism is: the maximum allowable dynamic fluctuation is regarded as a recoverable power resource, and the system will reallocate this part of the power within the current cycle, giving priority to supplying pile positions with high actual expected power load and low thermal potential.

10. The intelligent power control system of new energy charging pile based on artificial intelligence is characterized by: The system comprises: A user behavior credibility modeling module is used to obtain user-side data, environmental context, and pile position status information, combine them into a feature vector of the current pile position, input the feature vector into a neural network, output the probability of the user actually appearing and starting charging in the current scheduling period of the corresponding pile position and the estimated real power load value, and determine the corresponding actual expected power load, which is the weighted product of the probability of the user actually appearing and starting charging in the current scheduling period of the corresponding pile position and the estimated real power load value; A spatial thermal potential map construction module, used to embed the actual expected power load into the graph structure to perform thermal potential diffusion analysis and obtain the thermal potential value of the corresponding node; The resource scheduling optimization module is used to design a resource scheduling optimization function based on thermal potential diffusion analysis. The resource scheduling optimization function determines the actual available power allocation value for each pile position according to the thermal potential value and constraint conditions. The constraint conditions include that the total power scheduling of the entire station shall not exceed the system available value, the allocated power of each pile position shall not exceed its access physical limit, and the system will no longer allocate power to pile positions with extremely low confidence, so as to avoid waste of resources due to behavior prediction errors. The control instruction execution and feedback module is used to send the power allocation value actually available for each charging pile position to the actual charging pile control device, drive it to execute power output behavior according to the system strategy during the current scheduling cycle, and monitor the actual control execution status of each charging pile position in real time, and make dynamic adjustments based on the actual execution status.

Images