An artificial intelligence-based station-level electric vehicle charging load balancing method and system

Abstract

The application discloses a kind of station level electric vehicle charging load balancing method and system based on artificial intelligence, it is related to the field of automobile charging, its method includes: constructing N historical demand sample sequence;Training charging demand prediction model;Construct N current demand feature sequence;Output the predicted charging power sequence of N charging equipment;The corresponding predicted charging power sequence of N charging equipment is converted into phase;Construct station predicted power sequence;According to the time sequence of future scheduling period, construct power gap sequence;Along the arrangement direction of power gap sequence, determine target scheduling period and target power gap vector;According to target power gap vector, adjust target charging power, and output charging power control instruction;The application enables the charging power of multiple charging equipment to be coordinated according to the three-phase power constraint of station, realizes the balanced regulation of charging load in the range of station level.

Description

Technical Field

[0001] This invention relates to the field of vehicle charging, specifically to a site-level electric vehicle charging load balancing method and system based on artificial intelligence. Background Technology

[0002] With the continuous increase in the number of electric vehicles, the number of charging devices connected to electric vehicle charging stations at the same time is also increasing. Due to the differences in the access time, charging status, and charging demand of different charging devices, the simultaneous operation of multiple charging devices may cause the load to be concentrated in local time periods of the station. Especially in three-phase power supply scenarios, the power distribution of different phases corresponding to different charging devices can easily cause the power of a certain phase to exceed the allowable operating range.

[0003] Existing technologies typically adjust the load based on the currently collected operating power of the charging equipment, for example, reducing the charging power of some charging equipment when the current total load is detected to be too high. However, this type of method mainly responds to the current load state and cannot predict changes in charging power over future scheduling periods, causing load adjustments to usually occur after power limits have been exceeded. Furthermore, while existing technologies assist charging scheduling by predicting charging demand, they typically only allocate power based on the predicted total charging demand, without fully considering the three-phase power variation relationships formed when multiple charging equipment are connected to different phases. When the predicted charging demand of multiple charging equipment is concentrated in the same phase, even if the overall power of the station does not exceed the limit, single-phase power over-limit problems may still occur. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a site-level electric vehicle charging load balancing method and system based on artificial intelligence, which solves the technical problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention discloses a site-level electric vehicle charging load balancing method based on artificial intelligence, comprising the following steps: S1. Construct N historical demand sample sequences corresponding to N charging devices; S2. Train a charging demand prediction model based on N historical demand sample sequences; S3. Construct N current demand feature sequences corresponding to N charging devices within the current scheduling period; S4. Input the N current demand feature sequences into the charging demand prediction model, and output the predicted charging power sequence of each of the N charging devices in the K consecutive future scheduling periods. S5. Perform phase-by-phase conversion on the predicted charging power sequences corresponding to N charging devices to generate N phase-by-phase predicted power sequences; S6. Sum the N phase-specific predicted powers corresponding to the same future scheduling period in phase to construct the power station predicted power sequence. S7. Based on the predicted power sequence of the power station and the upper limit of the three-phase allowable power, generate power gap vectors corresponding to K future scheduling periods, and construct a power gap sequence according to the time order of the future scheduling periods. S8. Determine the target scheduling period with power gap and the corresponding target power gap vector along the arrangement direction of the power gap sequence; S9. Adjust the target charging power of the corresponding charging equipment within the target scheduling period according to the target power gap vector, and output the charging power control command.

[0006] In some embodiments, constructing N historical demand sample sequences corresponding to N charging devices includes: S1-1. Collect the multi-dimensional charging features of N charging devices in the charging station at M consecutive historical collection timestamps, and the historical charging power corresponding to K tag timestamps determined according to the preset prediction step size after each historical collection timestamp. The tag timestamp represents a future historical time point determined by using the corresponding historical collection timestamp as a benchmark and following a preset prediction step size.

[0007] S1-2. For any charging device, pair the multi-dimensional charging features corresponding to each historical collection timestamp with the historical charging power of the corresponding K tag timestamps to generate M historical demand samples for the corresponding charging device. S1-3. Arrange the M historical demand samples of the same charging device according to the time order of the M historical collection timestamps to construct the historical demand sample sequence of the corresponding charging device. S1-4. Traverse the N charging devices and repeatedly generate and arrange the historical demand samples until a sequence of N historical demand samples corresponding to the N charging devices is constructed.

[0008] In some embodiments, training the charging demand prediction model based on N historical demand sample sequences includes: S2-1. Input N historical demand sample sequences into the time series prediction model to be trained, and extract the charging state time series features corresponding to each historical demand sample sequence. S2-2, Calculate the attention weight of the temporal features of each charging state in the corresponding historical demand sample sequence; S2-3. Based on the attention weight, the charging state time sequence features in the same historical demand sample sequence are weighted and fused to generate the charging demand context vector of the corresponding charging device. S2-4. Input the charging demand context vectors corresponding to N charging devices into the power prediction layer of the time-series prediction model to be trained, and output the K training prediction charging power corresponding to each historical demand sample. S2-5. Compare the K training predicted charging power corresponding to each historical demand sample with the historical charging power corresponding to the K label timestamps, and calculate the charging power prediction loss. S2-6. Update the model parameters of the time series prediction model to be trained based on the charging power prediction loss, and repeat S2-1 to S2-5 to obtain the charging demand prediction model.

[0009] In some embodiments, constructing N current demand feature sequences corresponding to N charging devices within the current scheduling period includes: S3-1. Collect the real-time multi-dimensional charging characteristics of N charging devices within the current scheduling period, each at M consecutive real-time collection timestamps. S3-2. Arrange the M real-time multidimensional charging features of the same charging device according to the time order of the M real-time collection timestamps, and construct the current demand feature sequence of the corresponding charging device. S3-3. Traverse N charging devices and repeatedly perform the permutation of real-time multidimensional charging features until N current demand feature sequences corresponding to N charging devices are constructed. The real-time multidimensional charging features have the same feature dimensions and dimension arrangement order as the multidimensional charging features in the historical demand samples.

[0010] In some embodiments, the step of performing phase-by-phase conversion on the predicted charging power sequences corresponding to N charging devices to generate N phase-by-phase predicted power sequences includes: S5-1. In the predicted charging power sequence corresponding to any charging device, read the predicted charging power corresponding to K future scheduling periods one by one. S5-2. Calculate the predicted charging power corresponding to the K future scheduling periods with the L1 power allocation ratio, L2 power allocation ratio and L3 power allocation ratio of the corresponding charging equipment to generate K phase-by-phase predicted power vectors. The phase prediction power vector includes L1 prediction power, L2 prediction power and L3 prediction power corresponding to the future scheduling period.

[0011] S5-3. Arrange the K phase-by-phase predicted power vectors corresponding to the same charging device according to the time order of K future scheduling periods, and construct the phase-by-phase predicted power sequence of the corresponding charging device. S5-4. Traverse N charging devices and repeatedly execute the generation and arrangement of phase-by-phase predicted power vectors until N phase-by-phase predicted power sequences corresponding to N charging devices are constructed.

[0012] In some embodiments, constructing the predicted power sequence for power stations includes: S6-1. During the k-th future scheduling period, read the N phase prediction power vectors corresponding to the k-th future scheduling period from the N phase prediction power sequences respectively; S6-2. Sum the L1 predicted power, L2 predicted power and L3 predicted power in the N phase prediction power vectors respectively to generate the station prediction power vector corresponding to the k-th future scheduling period. S6-3. Traverse the K future scheduling periods, repeatedly execute phase-by-phase predicted power vector reading and in-phase summation until the K station predicted power vectors are obtained. S6-4. Arrange the predicted power vectors of the K power stations according to the time order of the K future scheduling periods to construct the predicted power sequence of the power stations.

[0013] In some embodiments, generating a power gap sequence based on the predicted power sequence of the power plant and the three-phase allowable power limit includes: S7-1. Read the kth predicted power vector of a power station from the predicted power sequence of the power stations; S7-2. Take the total predicted power of L1, L2 and L3 in the predicted power vector of the kth power station respectively, and calculate the difference between them and the corresponding allowed power upper limit of L1, allowed power upper limit of L2 and allowed power upper limit of L3 to generate a three-phase power difference vector. S7-3. Assign zero to the power difference of each phase that is not greater than zero in the three-phase power difference vector, and retain the power difference of each phase that is greater than zero to generate the power gap vector corresponding to the k-th future scheduling period. S7-4. Traverse the predicted power vectors of K power stations, repeatedly perform the subtraction of the corresponding phases and set the non-positive difference values ​​to zero, until K power gap vectors are obtained, and construct the power gap sequence according to the time order of future scheduling periods.

[0014] In some embodiments, determining the target scheduling period and the corresponding target power gap vector along the power gap sequence includes: S8-1. Search along the arrangement direction of the power gap sequence and mark the power gap vector that first contains a non-zero power gap component as the target power gap vector. S8-2, Mark the future scheduling period corresponding to the target power gap vector as the target scheduling period; S8-3. If none of the power gap vectors in the power gap sequence contain a non-zero power gap component, then the current scheduling cycle ends.

[0015] In some embodiments, adjusting the target charging power of the corresponding charging device within the target scheduling period according to the target power gap vector includes: S9-1. Obtain the predicted charging power of each charging device within the target scheduling period. S9-2. Determine the target charging power within the target scheduling period based on the target power gap vector and the adjustable power corresponding to each charging device. S9-3. Convert the target charging power into a charging power control command and send it to the corresponding charging device according to the device identifier.

[0016] This invention provides a site-level electric vehicle charging load balancing method and system based on artificial intelligence, which has the following beneficial effects: This invention constructs a historical demand sample sequence by collecting the charging characteristics of charging equipment over a continuous historical period and the historical charging power at corresponding future time points. It then uses a charging demand prediction model to output a predicted charging power sequence for multiple charging devices during future scheduling periods. This enables charging stations to perform advance analysis based on future charging load changes without having to passively adjust based solely on the current power status. Furthermore, the present invention performs phase-by-phase conversion on the predicted charging power corresponding to each charging device according to the corresponding three-phase power allocation ratio, and summarizes the predicted power of each phase of multiple charging devices in the same future scheduling period to form a station predicted power sequence, thereby enabling the future power change in different phases to be characterized separately, avoiding the problem of inaccurate judgment of single-phase power over-limit due to the phase difference of multiple charging devices accessing the station. Furthermore, the present invention generates a power gap sequence corresponding to future scheduling periods by comparing the predicted power sequence of the power station with the upper limit of the allowable power of each phase, and determines the target scheduling period in which the power gap first appears from multiple future scheduling periods, so that the power regulation process can handle specific over-limit periods and corresponding phase gaps. Furthermore, the present invention adjusts the target charging power of the charging equipment within the target scheduling period according to the target power gap vector, and generates corresponding charging power control commands, so that the charging power of multiple charging equipment can be coordinated and changed according to the three-phase power constraints of the station, thereby achieving balanced adjustment of the charging load within the station-level range.

[0017] Secondly, the present invention discloses an artificial intelligence-based depot-level electric vehicle charging load balancing system for executing the artificial intelligence-based depot-level electric vehicle charging load balancing method described in the first aspect, comprising: The sample sequence construction module is used to construct N historical demand sample sequences corresponding to N charging devices; The demand model training module is used to train a charging demand prediction model based on N historical demand sample sequences. The current sequence construction module is used to construct N current demand feature sequences corresponding to N charging devices within the current scheduling period; The demand sequence prediction module is used to input N current demand feature sequences into the charging demand prediction model and output the predicted charging power sequence of each of the N charging devices in K consecutive future scheduling periods. The phase sequence conversion module is used to perform phase conversion on the predicted charging power sequence corresponding to N charging devices to generate N phase predicted power sequences. The in-phase sequence construction module is used to sum the N phase-predicted powers corresponding to the same future scheduling period in the same phase to construct the power station prediction power sequence. The gap sequence construction module is used to generate power gap vectors corresponding to K future scheduling periods based on the power station's predicted power sequence and the three-phase allowable power limit, and to construct the power gap sequence according to the time order of the future scheduling periods; The dual-target determination module is used to determine the target scheduling period with power gap and the corresponding target power gap vector along the arrangement direction of the power gap sequence; The control command output module is used to adjust the target charging power of the corresponding charging equipment within the target scheduling period according to the target power gap vector, and output the charging power control command.

[0018] Compared with the prior art, the beneficial effects of the artificial intelligence-based electric vehicle charging load balancing system provided by the present invention are the same as those of the artificial intelligence-based electric vehicle charging load balancing method described above, so they will not be repeated here. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the station-level electric vehicle charging load balancing method based on artificial intelligence according to the present invention. Figure 2 This is a structural block diagram of the station-level electric vehicle charging load balancing system based on artificial intelligence according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: Please refer to Figure 1 This invention provides an artificial intelligence-based method for load balancing of electric vehicle charging stations, comprising the following steps: S1. Construct N historical demand sample sequences corresponding to N charging devices; Specifically, the historical demand sample sequence is used to characterize the charging state changes of a single charging device over a continuous historical time range.

[0022] The historical demand sample sequence is not directly obtained by arranging a single historical power data. Instead, it associates the device operating status corresponding to the historical time point with the actual charging power corresponding to the future time point, so that each historical collection time point corresponds to a future power change label.

[0023] Specifically, in this embodiment, step S1 includes: S1-1. Collect the multi-dimensional charging features of N charging devices in the charging station at M consecutive historical collection timestamps, and the historical charging power corresponding to K tag timestamps determined according to the preset prediction step size after each historical collection timestamp. The multi-dimensional charging features are used to describe the operating status of the charging device at the corresponding historical data collection timestamp, including but not limited to device operating parameters such as current charging power, current charging status, remaining charging demand, connection status, and charging duration.

[0024] The tag timestamp represents a future historical time point determined by using the corresponding historical collection timestamp as a benchmark and following a preset prediction step size.

[0025] For example, if the historical collection timestamp is 10:00 and the prediction step size is set to 5 minutes, the corresponding tag timestamps can include: 10:05, 10:10, 10:15.

[0026] Among them, the historical charging power corresponding to the tag timestamp is used to represent the actual change in charging power after a preset time interval under the state corresponding to the historical collection timestamp.

[0027] S1-2. For any charging device, pair the multi-dimensional charging features corresponding to each historical collection timestamp with the historical charging power of the corresponding K tag timestamps to generate M historical demand samples for the corresponding charging device. Specifically, each historical demand sample includes an input feature corresponding to a historical collection timestamp, and the actual charging power at multiple future time points corresponding to that input feature.

[0028] For example, the multi-dimensional charging features collected by a charging device at 10:00 are associated with the historical charging power corresponding to the three timestamps of 10:05, 10:10, and 10:15 to form a historical demand sample containing predictions of future targets at multiple time points.

[0029] S1-3. Arrange the M historical demand samples of the same charging device according to the time order of the M historical collection timestamps to construct the historical demand sample sequence of the corresponding charging device. Specifically, by arranging the historical collection timestamps from earliest to latest, multiple historical demand samples corresponding to the same charging device maintain a temporal continuity.

[0030] Each sample in the historical demand sample sequence corresponds to a continuous historical time position, which is used to describe the demand changes of charging equipment under different time states.

[0031] S1-4. Traverse the N charging devices and repeatedly generate and arrange the historical demand samples until a sequence of N historical demand samples corresponding to the N charging devices is constructed.

[0032] Specifically, different charging devices construct independent historical demand sample sequences, so that the historical charging state changes of each charging device can participate in model training.

[0033] S2. Train a charging demand prediction model based on N historical demand sample sequences; Specifically, the charging demand forecasting model is used to predict the charging power that the corresponding charging equipment may generate in multiple future scheduling periods based on the input continuous charging state change information.

[0034] In this embodiment, the charging demand prediction model adopts a time-series prediction model, which learns the correspondence between input features in historical demand sample sequences and future charging power to achieve multi-time point charging power prediction.

[0035] Specifically, in this embodiment, step S2 includes: S2-1. Input N historical demand sample sequences into the time series prediction model to be trained, and extract the charging state time series features corresponding to each historical demand sample sequence. Among them, the charging state time series feature is used to characterize the changing relationship between multiple consecutive time positions in the historical demand sample sequence.

[0036] For example, information such as the charging state change trend and power change trend corresponding to multiple consecutive time positions can be used as time series features extracted by time series prediction models.

[0037] S2-2, Calculate the attention weight of the temporal features of each charging state in the corresponding historical demand sample sequence; The formula for calculating the attention weight is: ; in: The attention weight represents the charging state temporal feature corresponding to the i-th historical time position; The importance score represents the charging state time-series feature corresponding to the i-th historical time position; M represents the number of historical demand samples in the historical demand sample sequence; j represents the j-th historical time position in the historical demand sample sequence.

[0038] Furthermore, the importance score It is obtained by linearly mapping the timing features of the i-th charging state, and the calculation formula is: ; in: This represents the timing feature of the charging state corresponding to the i-th historical time position; This represents the trainable parameter matrix corresponding to the attention weight calculation; This represents the bias parameter corresponding to the attention weight calculation.

[0039] Specifically, by assigning weights to the time-series features of charging status at different time points in the historical demand sample sequence, the model is able to distinguish the degree of influence of different historical time points on future changes in charging power.

[0040] S2-3. Based on the attention weight, the charging state time sequence features in the same historical demand sample sequence are weighted and fused to generate the charging demand context vector of the corresponding charging device. Among them, the charging demand context vector is used to comprehensively represent the overall demand change status of the corresponding charging device within the current historical time range.

[0041] S2-4. Input the charging demand context vectors corresponding to N charging devices into the power prediction layer of the time-series prediction model to be trained, and output the K training prediction charging power corresponding to each historical demand sample. Specifically, the power prediction layer outputs predicted power values ​​for multiple future time points based on the charging demand context vector, enabling the model to predict multiple future scheduling periods simultaneously.

[0042] S2-5. Compare the K training predicted charging power corresponding to each historical demand sample with the historical charging power corresponding to the K label timestamps, and calculate the charging power prediction loss. S2-6. Update the model parameters of the time series prediction model to be trained based on the charging power prediction loss, and repeat S2-1 to S2-5 to obtain the charging demand prediction model.

[0043] The training process of the above-mentioned time series model is based on iterative optimization of the model parameters by predictive loss, so that the time series prediction model to be trained gradually forms a mapping relationship between historical charging state characteristics and future charging power. This training method is a commonly used time series prediction model training method in this field.

[0044] S3. Construct N current demand feature sequences corresponding to N charging devices within the current scheduling period; Specifically, the current demand feature sequence is used to represent the change process of the continuous real-time operating status of each charging device within the current scheduling cycle.

[0045] Among them, the real-time multidimensional charging features in the current demand feature sequence maintain the same data dimensions and arrangement order as the multidimensional charging features in the historical demand sample sequence, so that the current state data can be input into the charging demand prediction model in accordance with the data structure consistent with the model training stage.

[0046] Specifically, in this embodiment, step S3 includes: S3-1. Collect the real-time multi-dimensional charging characteristics of N charging devices within the current scheduling period, each at M consecutive real-time collection timestamps. Among them, the real-time multidimensional charging feature is used to describe the real-time operating status of the charging equipment within the current scheduling cycle.

[0047] The dimensions of the real-time multidimensional charging features collected correspond to the multidimensional charging features in the historical demand samples, such as current charging power, current charging status, and remaining charging demand.

[0048] S3-2. Arrange the M real-time multidimensional charging features of the same charging device according to the time order of the M real-time collection timestamps, and construct the current demand feature sequence of the corresponding charging device. Specifically, by arranging the real-time data from earliest to latest according to the real-time collection timestamps, multiple real-time status data corresponding to the same charging device are arranged into a continuous time structure.

[0049] S3-3. Traverse N charging devices and repeatedly perform the permutation of real-time multidimensional charging features until N current demand feature sequences corresponding to N charging devices are constructed.

[0050] S4. Input the N current demand feature sequences into the charging demand prediction model, and output the predicted charging power sequence of each of the N charging devices in the K consecutive future scheduling periods. Specifically, the current demand feature sequence formed within the current scheduling period is input into the already trained charging demand prediction model, so that the model can predict the charging power corresponding to multiple future scheduling periods based on the current charging status change trend.

[0051] The predicted charging power sequence includes multiple predicted charging powers arranged according to future scheduling periods, with each predicted charging power corresponding to a future scheduling period.

[0052] For example, if the future scheduling periods include T1, T2, ..., TK, then the predicted charging power sequence can be represented as: ; in, This represents the predicted charging power of the i-th charging device in the k-th future scheduling period.

[0053] S5. Perform phase-by-phase conversion on the predicted charging power sequences corresponding to N charging devices to generate N phase-by-phase predicted power sequences; Specifically, since different charging devices may have different power supply phases when connected to the charging station, it is necessary to convert the single predicted charging power into the corresponding three-phase power distribution.

[0054] Among them, the phase-by-phase predicted power sequence is used to represent the predicted power changes of a single charging device in the three phases L1, L2, and L3 respectively during multiple future scheduling periods.

[0055] Specifically, in this embodiment, step S5 includes: S5-1. In the predicted charging power sequence corresponding to any charging device, read the predicted charging power corresponding to K future scheduling periods one by one. Specifically, the predicted charging power is read in the order of future scheduling periods, so that the predicted power can be associated with the corresponding time and location.

[0056] S5-2. Calculate the predicted charging power corresponding to the K future scheduling periods with the L1 power allocation ratio, L2 power allocation ratio and L3 power allocation ratio of the corresponding charging equipment to generate K phase-by-phase predicted power vectors. Wherein, the L1 power allocation ratio, L2 power allocation ratio, and L3 power allocation ratio represent the power proportion of the predicted charging power of the corresponding charging device in the three power supply phases, respectively.

[0057] S5-3. Arrange the K phase-by-phase predicted power vectors corresponding to the same charging device according to the time order of K future scheduling periods, and construct the phase-by-phase predicted power sequence of the corresponding charging device. Maintaining the order of future scheduling periods enables multiple three-phase power prediction states corresponding to the same charging equipment to form a time series.

[0058] S5-4. Traverse N charging devices and repeatedly execute the generation and arrangement of phase-by-phase predicted power vectors until N phase-by-phase predicted power sequences corresponding to N charging devices are constructed.

[0059] S6. Sum the N phase-specific predicted powers corresponding to the same future scheduling period in phase to construct the power station predicted power sequence. Specifically, the predicted power sequence of the charging station is used to represent the changes in the three-phase power of the charging station as a whole in multiple future scheduling periods.

[0060] Among them, the phase-specific predicted power sequence of a single charging device only describes the three-phase power contribution of the device itself, while the site-specific predicted power sequence summarizes the power of the same phase of multiple charging devices in the same future scheduling period.

[0061] Specifically, in this embodiment, step S6 includes: S6-1. During the k-th future scheduling period, read the N phase prediction power vectors corresponding to the k-th future scheduling period from the N phase prediction power sequences respectively; S6-2. Sum the L1 predicted power, L2 predicted power and L3 predicted power in the N phase prediction power vectors respectively to generate the station prediction power vector corresponding to the k-th future scheduling period. Specifically, for the k-th future scheduling period, the L1 predicted power corresponding to all charging devices is accumulated to obtain the total L1 predicted power for that period; similarly, the total L2 predicted power and the total L3 predicted power are obtained.

[0062] S6-3. Traverse the K future scheduling periods, repeatedly execute phase-by-phase predicted power vector reading and in-phase summation until the K station predicted power vectors are obtained. S6-4. Arrange the predicted power vectors of the K power stations according to the time order of the K future scheduling periods to construct the predicted power sequence of the power stations.

[0063] S7. Based on the predicted power sequence of the power station and the upper limit of the three-phase allowable power, generate power gap vectors corresponding to K future scheduling periods, and construct a power gap sequence according to the time order of the future scheduling periods. Specifically, the power gap vector is used to represent the extent to which the predicted power of each phase exceeds the corresponding allowable power limit during a future scheduling period.

[0064] Specifically, only the power difference exceeding the allowable power limit is retained, while the phase difference corresponding to the phase that does not exceed the allowable power limit is set to zero.

[0065] Specifically, in this embodiment, step S7 includes: S7-1. Read the kth predicted power vector of a power station from the predicted power sequence of the power stations; S7-2. Take the total predicted power of L1, L2 and L3 in the predicted power vector of the kth power station respectively, and calculate the difference between them and the corresponding allowed power upper limit of L1, allowed power upper limit of L2 and allowed power upper limit of L3 to generate a three-phase power difference vector. Among them, the three-phase power difference vector is used to characterize the degree of deviation of the predicted power of each phase from the corresponding allowable power limit in the k-th future scheduling period.

[0066] Specifically, for any phase, the predicted total power is subtracted from the corresponding allowable power limit. If the result is positive, it means that the predicted power of that phase exceeds the allowable range; if the result is negative, it means that the phase is still within the allowable range.

[0067] S7-3. Assign zero to the power difference of each phase that is not greater than zero in the three-phase power difference vector, and retain the power difference of each phase that is greater than zero to generate the power gap vector corresponding to the k-th future scheduling period. It should be noted that setting the difference that is not greater than zero to zero is to retain only the phase information that needs to be regulated by power, so that the non-zero dimensions in the power gap vector correspond to the phases that have the risk of power over-limit.

[0068] For example, if the three-phase power difference vector corresponding to a certain future scheduling period is (−20,50,−10), then after setting the non-positive difference value to zero, it generates (0,50,0). The second dimension indicates a power gap of 50kW in the corresponding phase.

[0069] S7-4. Traverse the predicted power vectors of K power stations, repeatedly perform the corresponding phase difference calculation and set non-positive difference values ​​to zero until K power gap vectors are obtained, and construct the power gap sequence according to the time order of future scheduling periods.

[0070] Furthermore, the power gap sequence organizes the power over-limit states corresponding to multiple future scheduling periods in chronological order, so that each future scheduling period has a corresponding power gap description.

[0071] Specifically, by converting the future power forecast state into a power deficit state, the power station can identify power constraint deviations on specific phases from the predicted power changes.

[0072] S8. Determine the target scheduling period with power gap and the corresponding target power gap vector along the arrangement direction of the power gap sequence; Specifically, by traversing the power gap sequence according to the future scheduling period order, the future scheduling period in which the earliest power over-limit risk occurs is determined first.

[0073] Specifically, in this embodiment, step S8 includes: S8-1. Search along the arrangement direction of the power gap sequence and mark the power gap vector that first contains a non-zero power gap component as the target power gap vector. Specifically, the process is to traverse the power gap sequence according to the order of future scheduling periods and determine whether each power gap vector has a non-zero dimension.

[0074] Among them, the non-zero power gap component indicates that at least one phase in the corresponding future scheduling period exceeds the allowable power limit, and power regulation needs to be performed.

[0075] S8-2, Mark the future scheduling period corresponding to the target power gap vector as the target scheduling period; The target scheduling period is not fixed to the first scheduling period after the current time, but is determined based on the power gap status corresponding to multiple future scheduling periods, so that the power adjustment object corresponds to the actual predicted over-limit time.

[0076] S8-3. If none of the power gap vectors in the power gap sequence contain a non-zero power gap component, then the current scheduling cycle ends.

[0077] Furthermore, if there is no situation where the power exceeds the allowable power limit throughout the entire prediction time range, no power adjustment object is generated, and the current charging equipment operating state is maintained.

[0078] Specifically, by performing time-sequence detection on the power gap sequence, the target time position and corresponding phase gap that need to be adjusted for load can be determined.

[0079] S9. Adjust the target charging power of the corresponding charging equipment within the target scheduling period according to the target power gap vector, and output the charging power control command. Specifically, after determining the target scheduling period, the power of charging equipment that can reduce charging power is adjusted according to the power gap corresponding to that period.

[0080] Specifically, in this embodiment, step S9 includes: S9-1. Obtain the predicted charging power of each charging device within the target scheduling period. Since the power gap corresponding to the target scheduling period comes from the sum of the predicted power of multiple charging devices, it is necessary to read the predicted charging power of each charging device at that time position as the initial value for power adjustment.

[0081] S9-2. Determine the target charging power within the target scheduling period based on the target power gap vector and the adjustable power corresponding to each charging device. Among them, adjustable power indicates the range of charging power that the charging device can reduce while meeting its minimum charging requirements.

[0082] Specifically, for any charging device, its adjustable power can be determined based on the difference between the predicted charging power and the minimum allowable charging power corresponding to the target scheduling period.

[0083] Furthermore, after determining the adjustable power of multiple charging devices, based on the phase gap situation corresponding to the target power gap vector, the charging device that can reduce the power of the corresponding phase is selected, and its target charging power is adjusted.

[0084] S9-3. Convert the target charging power into a charging power control command and send it to the corresponding charging device according to the device identifier.

[0085] Specifically, the target charging power represents the power value that the charging device needs to perform during the target scheduling period. The charging power control command is used to convert this target power value into control parameters that the charging device can recognize.

[0086] It should be noted that the control command carries the corresponding device identifier, enabling different charging devices to receive power adjustment parameters corresponding to themselves.

[0087] This embodiment converts the predicted power gap into the corresponding power adjustment amount of the charging equipment, so that the overall power status of the power station can be adjusted within the three-phase allowable power range.

[0088] This invention uses time-series modeling of the historical operating status of charging equipment to predict the charging power of multiple charging equipment during future scheduling periods. Furthermore, it combines the three-phase power distribution of the charging equipment to analyze the future three-phase power status of the charging station. When a future power gap occurs, the corresponding scheduling period is determined and a charging power control command is generated. This enables the charging station to make advance adjustments for possible phase power over-limit situations in the future, thereby achieving coordinated load balancing among multiple charging equipment.

[0089] This invention also provides an artificial intelligence-based site-level electric vehicle charging load balancing system, which is used to implement the above-described method embodiments; details already described will not be repeated. The terms "module," "unit," and "subunit," etc., used below refer to combinations of software and / or hardware that perform predetermined functions. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0090] Figure 2 This is a structural block diagram of the station-level electric vehicle charging load balancing system based on artificial intelligence according to the present invention. The system includes: The sample sequence construction module is used to construct N historical demand sample sequences corresponding to N charging devices; The demand model training module is used to train a charging demand prediction model based on N historical demand sample sequences. The current sequence construction module is used to construct N current demand feature sequences corresponding to N charging devices within the current scheduling period; The demand sequence prediction module is used to input N current demand feature sequences into the charging demand prediction model and output the predicted charging power sequence of each of the N charging devices in K consecutive future scheduling periods. The phase sequence conversion module is used to perform phase conversion on the predicted charging power sequence corresponding to N charging devices to generate N phase predicted power sequences. The in-phase sequence construction module is used to sum the N phase-predicted powers corresponding to the same future scheduling period in the same phase to construct the power station prediction power sequence. The gap sequence construction module is used to generate power gap vectors corresponding to K future scheduling periods based on the power station's predicted power sequence and the three-phase allowable power limit, and to construct the power gap sequence according to the time order of the future scheduling periods; The dual-target determination module is used to determine the target scheduling period with power gap and the corresponding target power gap vector along the arrangement direction of the power gap sequence; The control command output module is used to adjust the target charging power of the corresponding charging equipment within the target scheduling period according to the target power gap vector, and output the charging power control command.

[0091] It should be noted that each functional module in the above system corresponds to each step in the above method embodiment and is used to implement the same technical solution. The specific execution process of each functional module and the technical effect achieved can be found in the relevant description of the above method embodiment, and will not be repeated here.

[0092] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A site-level electric vehicle charging load balancing method based on artificial intelligence, characterized in that, include: S1. Construct N historical demand sample sequences corresponding to N charging devices; S2. Train a charging demand prediction model based on N historical demand sample sequences; S3. Construct N current demand feature sequences corresponding to N charging devices within the current scheduling period; S4. Input the N current demand feature sequences into the charging demand prediction model, and output the predicted charging power sequence of each of the N charging devices in the K consecutive future scheduling periods. S5. Perform phase-by-phase conversion on the predicted charging power sequences corresponding to N charging devices to generate N phase-by-phase predicted power sequences; The generation of N phase-specific predicted power sequences includes: S5-1. In the predicted charging power sequence corresponding to any charging device, read the predicted charging power corresponding to K future scheduling periods one by one. S5-2. Calculate the predicted charging power corresponding to the K future scheduling periods with the L1 power allocation ratio, L2 power allocation ratio and L3 power allocation ratio of the corresponding charging equipment to generate K phase-by-phase predicted power vectors. The phase prediction power vector includes L1 prediction power, L2 prediction power and L3 prediction power corresponding to the future scheduling period; S5-3. Arrange the K phase-by-phase predicted power vectors corresponding to the same charging device according to the time order of K future scheduling periods, and construct the phase-by-phase predicted power sequence of the corresponding charging device. S5-4. Traverse N charging devices and repeatedly execute the generation and arrangement of phase-by-phase predicted power vectors until N phase-by-phase predicted power sequences corresponding to N charging devices are constructed. S6. Sum the N phase-specific predicted powers corresponding to the same future scheduling period in phase to construct the power station predicted power sequence. The construction of the predicted power sequence for power plants includes: S6-1. During the k-th future scheduling period, read the N phase prediction power vectors corresponding to the k-th future scheduling period from the N phase prediction power sequences respectively; S6-2. Sum the L1 predicted power, L2 predicted power and L3 predicted power in the N phase prediction power vectors respectively to generate the station prediction power vector corresponding to the k-th future scheduling period. S6-3. Traverse the K future scheduling periods, repeatedly execute phase-by-phase predicted power vector reading and in-phase summation until the K station predicted power vectors are obtained. S6-4. Arrange the K predicted power vectors of the power stations according to the time order of the K future scheduling periods to construct the predicted power sequence of the power stations; S7. Based on the predicted power sequence of the power station and the upper limit of the three-phase allowable power, generate power gap vectors corresponding to K future scheduling periods, and construct a power gap sequence according to the time order of the future scheduling periods. The constructed power gap sequence includes: S7-1. Read the kth predicted power vector of a power station from the predicted power sequence of the power stations; S7-2. Take the total predicted power of L1, L2 and L3 in the predicted power vector of the kth power station respectively, and calculate the difference between them and the corresponding allowed power upper limit of L1, allowed power upper limit of L2 and allowed power upper limit of L3 to generate a three-phase power difference vector. S7-3. Assign zero to the power difference of each phase that is not greater than zero in the three-phase power difference vector, and retain the power difference of each phase that is greater than zero to generate the power gap vector corresponding to the k-th future scheduling period. S7-4. Traverse the predicted power vectors of K power stations, repeatedly perform the corresponding phase difference calculation and set non-positive difference values ​​to zero until K power gap vectors are obtained, and construct the power gap sequence according to the time order of future scheduling periods; S8. Determine the target scheduling period with power gap and the corresponding target power gap vector along the arrangement direction of the power gap sequence; S9. Adjust the target charging power of the corresponding charging equipment within the target scheduling period according to the target power gap vector, and output the charging power control command.

2. The site-level electric vehicle charging load balancing method based on artificial intelligence according to claim 1, characterized in that, The construction of N historical demand sample sequences corresponding to N charging devices includes: S1-1. Collect the multi-dimensional charging features of N charging devices in the charging station at M consecutive historical collection timestamps, and the historical charging power corresponding to K tag timestamps determined according to the preset prediction step size after each historical collection timestamp. The tag timestamp represents a future historical time point determined by using the corresponding historical collection timestamp as a benchmark and following a preset prediction step size. S1-2. For any charging device, pair the multi-dimensional charging features corresponding to each historical collection timestamp with the historical charging power of the corresponding K tag timestamps to generate M historical demand samples for the corresponding charging device. S1-3. Arrange the M historical demand samples of the same charging device according to the time order of the M historical collection timestamps to construct the historical demand sample sequence of the corresponding charging device. S1-4. Traverse the N charging devices and repeatedly generate and arrange the historical demand samples until a sequence of N historical demand samples corresponding to the N charging devices is constructed.

3. The site-level electric vehicle charging load balancing method based on artificial intelligence according to claim 1, characterized in that, The step of training a charging demand prediction model based on N historical demand sample sequences includes: S2-1. Input N historical demand sample sequences into the time series prediction model to be trained, and extract the charging state time series features corresponding to each historical demand sample sequence. S2-2, Calculate the attention weight of the temporal features of each charging state in the corresponding historical demand sample sequence; S2-3. Based on the attention weight, the charging state time sequence features in the same historical demand sample sequence are weighted and fused to generate the charging demand context vector of the corresponding charging device. S2-4. Input the charging demand context vectors corresponding to N charging devices into the power prediction layer of the time-series prediction model to be trained, and output the K training prediction charging power corresponding to each historical demand sample. S2-5. Compare the K training predicted charging power corresponding to each historical demand sample with the historical charging power corresponding to the K label timestamps, and calculate the charging power prediction loss. S2-6. Update the model parameters of the time series prediction model to be trained based on the charging power prediction loss, and repeat S2-1 to S2-5 to obtain the charging demand prediction model.

4. The site-level electric vehicle charging load balancing method based on artificial intelligence according to claim 1, characterized in that, The construction of N current demand feature sequences corresponding to N charging devices within the current scheduling period includes: S3-1. Collect the real-time multi-dimensional charging characteristics of N charging devices within the current scheduling period, each at M consecutive real-time collection timestamps. S3-2. Arrange the M real-time multidimensional charging features of the same charging device according to the time order of the M real-time collection timestamps, and construct the current demand feature sequence of the corresponding charging device. S3-3. Traverse N charging devices and repeatedly perform the permutation of real-time multidimensional charging features until N current demand feature sequences corresponding to N charging devices are constructed. The real-time multidimensional charging features have the same feature dimensions and dimension arrangement order as the multidimensional charging features in the historical demand samples.

5. The site-level electric vehicle charging load balancing method based on artificial intelligence according to claim 1, characterized in that, Determine the target scheduling period and the corresponding target power gap vector along the power gap sequence, including: S8-1. Search along the arrangement direction of the power gap sequence and mark the power gap vector that first contains a non-zero power gap component as the target power gap vector. S8-2, Mark the future scheduling period corresponding to the target power gap vector as the target scheduling period; S8-3. If none of the power gap vectors in the power gap sequence contain a non-zero power gap component, then the current scheduling cycle ends.

6. The site-level electric vehicle charging load balancing method based on artificial intelligence according to claim 1, characterized in that, The step of adjusting the target charging power of the corresponding charging equipment within the target scheduling period according to the target power gap vector includes: S9-1. Obtain the predicted charging power of each charging device within the target scheduling period. S9-2. Determine the target charging power within the target scheduling period based on the target power gap vector and the adjustable power corresponding to each charging device. S9-3. Convert the target charging power into a charging power control command and send it to the corresponding charging device according to the device identifier.

7. An AI-based depot-level electric vehicle charging load balancing system, used to execute the AI-based depot-level electric vehicle charging load balancing method according to any one of claims 1 to 6, characterized in that, include: The sample sequence construction module is used to construct N historical demand sample sequences corresponding to N charging devices; The demand model training module is used to train a charging demand prediction model based on N historical demand sample sequences. The current sequence construction module is used to construct N current demand feature sequences corresponding to N charging devices within the current scheduling period; The demand sequence prediction module is used to input N current demand feature sequences into the charging demand prediction model and output the predicted charging power sequence of each of the N charging devices in K consecutive future scheduling periods. The phase sequence conversion module is used to perform phase conversion on the predicted charging power sequence corresponding to N charging devices to generate N phase predicted power sequences. The generation of N phase-specific predicted power sequences includes: S5-1. In the predicted charging power sequence corresponding to any charging device, read the predicted charging power corresponding to K future scheduling periods one by one. S5-2. Calculate the predicted charging power corresponding to the K future scheduling periods with the L1 power allocation ratio, L2 power allocation ratio and L3 power allocation ratio of the corresponding charging equipment to generate K phase-by-phase predicted power vectors. The phase prediction power vector includes L1 prediction power, L2 prediction power and L3 prediction power corresponding to the future scheduling period; S5-3. Arrange the K phase-by-phase predicted power vectors corresponding to the same charging device according to the time order of K future scheduling periods, and construct the phase-by-phase predicted power sequence of the corresponding charging device. S5-4. Traverse N charging devices and repeatedly execute the generation and arrangement of phase-by-phase predicted power vectors until N phase-by-phase predicted power sequences corresponding to N charging devices are constructed. The in-phase sequence construction module is used to sum the N phase-predicted powers corresponding to the same future scheduling period in the same phase to construct the power station prediction power sequence. The construction of the predicted power sequence for power plants includes: S6-1. During the k-th future scheduling period, read the N phase prediction power vectors corresponding to the k-th future scheduling period from the N phase prediction power sequences respectively; S6-2. Sum the L1 predicted power, L2 predicted power and L3 predicted power in the N phase prediction power vectors respectively to generate the station prediction power vector corresponding to the k-th future scheduling period. S6-3. Traverse the K future scheduling periods, repeatedly execute phase-by-phase predicted power vector reading and in-phase summation until the K station predicted power vectors are obtained. S6-4. Arrange the K predicted power vectors of the power stations according to the time order of the K future scheduling periods to construct the predicted power sequence of the power stations; The gap sequence construction module is used to generate power gap vectors corresponding to K future scheduling periods based on the power station's predicted power sequence and the three-phase allowable power limit, and to construct the power gap sequence according to the time order of the future scheduling periods; The constructed power gap sequence includes: S7-1. Read the kth predicted power vector of a power station from the predicted power sequence of the power stations; S7-2. Take the total predicted power of L1, L2 and L3 in the predicted power vector of the kth power station respectively, and calculate the difference between them and the corresponding allowed power upper limit of L1, allowed power upper limit of L2 and allowed power upper limit of L3 to generate a three-phase power difference vector. S7-3. Assign zero to the power difference of each phase that is not greater than zero in the three-phase power difference vector, and retain the power difference of each phase that is greater than zero to generate the power gap vector corresponding to the k-th future scheduling period. S7-4. Traverse the predicted power vectors of K power stations, repeatedly perform the corresponding phase difference calculation and set non-positive difference values ​​to zero until K power gap vectors are obtained, and construct the power gap sequence according to the time order of future scheduling periods; The dual-target determination module is used to determine the target scheduling period with power gap and the corresponding target power gap vector along the arrangement direction of the power gap sequence; The control command output module is used to adjust the target charging power of the corresponding charging equipment within the target scheduling period according to the target power gap vector, and output the charging power control command.

Images