Building energy optimization scheduling method considering electric vehicle charging load uncertainty

Abstract

This invention discloses a building energy optimization scheduling method that considers the uncertainty of electric vehicle charging loads. The method comprises the following steps: establishing a day-ahead prediction model for electric vehicle charging loads; quantifying the uncertainty of charging loads based on the charging load prediction model; and establishing a building energy optimization scheduling model that considers the uncertainty of charging loads. The strategy proposed in this invention that considers the uncertainty of electric vehicle charging loads is more adaptable to various load conditions that may occur the next day, thereby improving the robustness of the scheduling strategy. Furthermore, the combined optimization scheduling of air conditioning and charging loads achieves complementarity within the scheduling time period, further optimizing load variance and stabilizing the operation of the building energy system.

Description

Technical Field

[0001] The present invention belongs to the field of building integrated energy systems, and in particular relates to a building energy optimization scheduling method considering the uncertainty of electric vehicle charging load. Background Art

[0002] With the widespread adoption of charging facilities in future buildings, the interaction between electric vehicles and buildings will become increasingly close. The integration of electric vehicles has both advantages and disadvantages. Excessive concentration of electric vehicles can lead to load fluctuations, imbalances in distribution system component capacity, voltage, and frequency, power loss, and instability in the distribution network. However, integrating electric vehicle charging scheduling with electricity market operations and rationally arranging charging can help mitigate grid fluctuations through demand response and ensure grid stability. Orderly charging of electric vehicles can help shift or reduce peak loads and reduce associated electricity costs. Building energy management system scheduling can shift peak loads and fill valleys, and by coordinating the operation of electric vehicles and distributed power sources, it can improve the local consumption of new energy. Therefore, comprehensively utilizing flexible loads such as charging loads to develop optimized scheduling strategies can significantly improve building operations.

[0003] However, current strategies based on deterministic models often overlook the impact of uncertainties in reality, and the resulting strategies are likely to be inappropriate for actual situations. Therefore, with the rapid development of electric vehicles, it is essential to study optimization methods for building integrated scheduling strategies under uncertain conditions of electric vehicle charging loads. This approach has both theoretical and practical value and will provide a theoretical basis for robust operational management of building energy systems, thereby aligning with the general trend of electric vehicle promotion, responding to the dual carbon goals, and promoting the resolution of energy and environmental issues. Summary of the Invention

[0004] In view of this, a building energy optimization scheduling method considering the uncertainty of electric vehicle charging load is proposed to further improve the robustness of the scheduling strategy. At the same time, the complementarity of the joint optimization scheduling of air-conditioning load and charging load in the scheduling time period is utilized to optimize the load variance and make the operation of the building energy system more stable.

[0005] To achieve the above objectives, a building energy optimization scheduling method considering the uncertainty of electric vehicle charging load is proposed, which includes the following steps:

[0006] Step 1): Establish a day-ahead forecast model for electric vehicle charging load

[0007] A questionnaire survey was conducted in a certain area to obtain actual electric vehicle load data. Based on the collected data, user commuting behavior characteristics and the distribution of electric vehicle physical properties were analyzed, and an electric vehicle charging load prediction model was established based on Monte Carlo simulation and statistical methods.

[0008] Step 2): Quantify charging load uncertainty

[0009] Based on the charging load prediction model, the Monte Carlo method introduces errors and makes corrections based on the expected values ​​of each parameter. The next day's charging load samples, obtained after a sufficient number of Monte Carlo simulations, are combined into a set to form a collection of all possible charging scenarios for the next day. The building's air conditioning load under pre-cooling conditions, as well as basic power loads such as lighting, office equipment, and elevators, are measured and combined with the charging load to form a collection of all possible power load scenarios for the next day.

[0010] Step 3): Establish a building energy optimization scheduling model considering charging load uncertainty

[0011] A single-objective genetic algorithm is used to optimize the most unfavorable load variance in all possible scenarios. The charging start time of each electric vehicle is used as a variable, and the arrival guarantee rate and charging guarantee rate are proposed as constraints to establish an optimal scheduling model for electric vehicle charging facilities.

[0012] The step 1) establishes a day-ahead prediction model for electric vehicle charging load, specifically:

[0013] (1) Obtain the distribution of relevant parameters through questionnaire survey

[0014] A questionnaire survey was conducted in a specific region to obtain actual electric vehicle load data. By analyzing the survey results, a probability density distribution function was obtained for indicators such as arrival time at work, departure time at work, distance traveled from home to work, SOC at departure, battery capacity, and power consumption per kilometer.

[0015] (2) Calculation of hourly charging load for a single vehicle

[0016] To simplify the calculation of charging load, it is assumed that the electric vehicle does not leave during operation and the charging process is considered to be constant power. According to national standards, the slow charging power is set at 7kW. This invention only establishes a full slow charging scenario for the configuration of charging pile types.

[0017] After obtaining samples of each random variable through Monte Carlo extraction, the SOC of each electric vehicle when it arrives at the station can be calculated:

[0018]

[0019] Where, SOC arr，n SOC is the state of charge when the nth vehicle arrives at the work site; de,n K is the state of charge of the nth car when it leaves home; nis the power consumption per kilometer of the nth vehicle, kWh / km; D n is the driving distance of the nth vehicle from home to work, km; Cap n is the battery capacity of the nth vehicle, kWh.

[0020] From this, the charging time for each vehicle can be calculated:

[0021]

[0022] Where, L n is the charging time of the nth vehicle, hours; P c，n is the charging power of the nth vehicle, in kW. In the full slow charging scenario, the charging power of all vehicles is 7 kW.

[0023] Assume that the charging state variable ε of the nth electric vehicle at a certain time t in one day is n (t):

[0024]

[0025] Where, t s,n is the moment when the nth vehicle starts charging, and the vehicle charging time ε n (t)=1, when not charging, it is ε n (t)=0.

[0026] The hourly charging load of a single vehicle and multiple vehicles is:

[0027] P EV,n (t) = P c,n ×ε n (t) (4)

[0028]

[0029] Where, P EV,n (t) is the charging load of the nth vehicle at time t, kW; P EV (t) is the charging load of all vehicles at time t, in kW; N is the total number of vehicles. This gives the hourly charging load of electric vehicles of a certain scale in a day.

[0030] (3) Monte Carlo simulation process

[0031] For the full slow charging scenario, starting from the first vehicle, randomly extract the start time of charging, electric vehicle battery capacity, power consumption per kilometer, distance from home to work, and SOC at departure according to the probability density distribution function of each random variable, and calculate the hourly charging load of the vehicle according to formulas (1) to (5). The remaining vehicles perform the same extraction and calculation, accumulate the hourly charging load, and complete one Monte Carlo simulation process. Repeat the above extraction process to obtain a sufficient number of samples, and perform a total of 10,000 extractions. The Monte Carlo extraction process for the full slow charging scenario is as follows: Figure 2 As shown. Figure 2 After the extraction is completed, 10,000 sets of hourly electric vehicle charging load sample data can be obtained. By statistically calculating the maximum, minimum, and median values ​​of the electric vehicle charging load at each moment, the distribution characteristics of the electric vehicle load can be obtained.

[0032] The step 2) quantifies the uncertainty of the charging load, specifically:

[0033] (1) Correction of expectations and extraction of errors

[0034] For each vehicle, in order to simulate the uncertain charging load prediction, the possible errors of the uncertain factors are extracted, and the expected values ​​of each factor are corrected to form a set of "situations that need to be considered".

[0035] The arrival time expression considering uncertainty deviation is:

[0036]

[0037] Where, t s,n (exp) is the expected value of the time when the nth vehicle arrives, i.e., the time when charging starts; The present invention uses the Monte Carlo method to generate a set of finite but sufficiently large error vectors:

[0038]

[0039] Where, is the error vector of the arrival time prediction value, which contains all possible prediction errors; is the error observation vector extracted by Monte Carlo simulation; is the predicted value of the arrival time of the ith extraction; Ne is the number of Monte Carlo extractions. When Ne is large enough, it can be considered that yes A fairly good approximation is given by setting Ne = 10000. And the prediction error follows a normal distribution:

[0040]

[0041] Where, is the standard deviation of the arrival time prediction error; is the expected error in the arrival time prediction.

[0042] Similarly, the expression of the departure time considering uncertainty deviation is:

[0043]

[0044] Where, t l,n (exp) is the expected value of the latest time when the nth vehicle leaves, i.e., the end of charging; is the prediction error value of the departure time; is the error vector of the predicted value at the departure time, which contains all possible prediction errors; is the error observation vector extracted by Monte Carlo simulation; is the predicted value of the departure time extracted for the i-th time; is the probability density distribution function of the departure time prediction error; is the standard deviation of the prediction error of the departure time; is the expected error in the prediction of the departure time.

[0045] Similarly, the expression of SOC when considering uncertainty is:

[0046]

[0047] Where, SOC arr,n (exp) is the expected value of SOC when the nth vehicle arrives; is the predicted error value of SOC at arrival; is the error vector of the predicted SOC value at arrival, which includes all possible prediction errors; is the error observation vector extracted by Monte Carlo simulation; is the predicted SOC value at arrival extracted for the i-th time; is the probability density distribution function of the SOC prediction error at arrival; is the standard deviation of the SOC prediction error at arrival; is the expected SOC prediction error upon arrival.

[0048] (2) Forming a set of all possible charging loads

[0049] After obtaining a calculation expression that takes into account the uncertainty of each variable, the Monte Carlo extraction method is used to extract parameters for all vehicles each time. The results of each extraction are used as a set of possible scenarios for the next day. This extraction is repeated 10,000 times. By calculating the charging load, 10,000 sets of hourly charging load prediction results are obtained. This set is used as the total possible charging load curve for the next day, as shown in the following formula:

[0050]

[0051] Where, Ω EV is the charging load vector, which includes all possible situations on the next day; is the charging load observation vector extracted by Monte Carlo simulation; is the charging load of electric vehicles in group i within one day; is the charging load of the electric vehicles in group i at time t.

[0052] (3) Forming a set of all possible total power loads

[0053]

[0054] P s (t) = A × I (t) × η d (18)

[0055]

[0056] Where, Net power load of group i at time t, kW; P AC (t) is the air conditioning load at time t, kW; P other (t) is the other building basic power load at time t, including lighting, equipment, elevators, etc., in kW; is the average value of the total power load of group i, kW; P s (t) is the photovoltaic output at time t, which is 0 in the no-PV scenario, kW; A is the area of ​​the photovoltaic panel, m 2 ; I(t) is the total radiation intensity on the photovoltaic panel at time t, W / m 2 ;η d is the photovoltaic conversion efficiency.

[0057] Wherein, the step 3) establishes a building energy optimization scheduling model considering the uncertainty of charging load, specifically:

[0058] (1) Determine the optimization variables

[0059] During the operational phase, the uncertainty in the EV charging load comes from arrival time and the SOC upon arrival. To ensure stable operation of the building energy system, uncertainty is introduced into the EV charging load. The next day's charging start time for each EV is used as the optimization variable. Based on the predicted charging load for the next day, the start time for each EV is determined.

[0060] (2) Determine the objective function

[0061] Since excessive load fluctuations will seriously affect the operational stability and safety of the power grid, the present invention uses the load variance of the total power load as the optimization target. Considering the uncertainty, various charging load curves may appear on the second day. Therefore, the final strategy obtained by optimization should optimize all possible scenarios on the second day as much as possible, that is, regardless of whether Ω EV In either case, the load variance should be improved after using the optimization strategy. The objective function is determined to be the minimization of the maximum load variance, and the calculation formula is as follows:

[0062]

[0063] Where Object is the objective function, which means minimizing the maximum value (the most unfavorable case) of the load variance in all scenarios; Net power load of group i at time t, kW; is the average value of the total power load of group i, kW.

[0064] (3) Determine the constraints

[0065] ① Electric vehicle starting charging time constraints

[0066] The earliest charging start time for a single electric vehicle is the time of arrival. It is assumed that the SOC ≥ 0.8 when the vehicle leaves the vehicle satisfies the owner's charging needs. Therefore, the latest charging start time plus the time required to charge to SOC = 0.8 cannot be later than the departure time of the vehicle.

[0067]

[0068] Where, is the minimum value among the 10,000 arrival times of the nth vehicle; t res,n The optimized charging start time for the nth vehicle; The time required for the nth vehicle in group i to charge to SOC = 0.8; It is the maximum value among the latest charging start times of the nth vehicle.

[0069] ② Arrival rate is not guaranteed

[0070] Since the arrival times of various vehicles on the next day vary, if the calculated charging start time for a vehicle on the next day is earlier than the vehicle's actual arrival time, the strategy is considered invalid for that vehicle and it is necessary to wait for the vehicle to arrive before re-assigning a charging start time on site, which reduces the effectiveness of the day-ahead optimization scheduling. To ensure the robustness of the strategy and adapt to various situations, the concept of arrival non-guarantee rate is proposed:

[0071]

[0072] Where, N is the non-guaranteed rate at the destination; notarrive is the total number of vehicles whose arrival time is later than the re-set charging start time in all scenarios.

[0073] The stipulation is that the arrival failure rate must not exceed 10%, meaning that the strategy is effective for 90% of vehicles in all scenarios. Since this arrival failure is very easy to occur, directly using hard constraints within a certain range may cause the optimization calculation to fail. Therefore, soft constraints are set to penalize the objective function when the conditions are not met:

[0074]

[0075] Where ∈1 is the slack variable and ≥0.

[0076] ③Charging rate constraint is not guaranteed

[0077] If a car charges according to the policy's reset charging start time and does not reach SOC = 0.8 when leaving, it is considered that the policy does not meet the owner's charging needs and this situation should be avoided. In order to ensure the robustness of the obtained policy and ensure that the policy can meet the user's charging needs as much as possible in various scenarios, the present invention proposes the concept of charging non-guarantee rate:

[0078]

[0079] Where, The charging rate is not guaranteed; N notfull The total number of vehicles whose SOC does not reach 0.8 when leaving in all scenarios.

[0080] Just like the arrival guarantee, the objective function is penalized when the conditions are not met:

[0081]

[0082] Where ∈2 is a slack variable, ≥0.

[0083] Beneficial effects

[0084] (1) Compared with the traditional deterministic strategy, the strategy proposed in this paper that considers the uncertainty of electric vehicle charging load is more adaptable to various load conditions that may occur the next day, thereby improving the robustness of the scheduling strategy.

[0085] (2) The joint optimization scheduling of air-conditioning load and charging load achieves complementarity in the scheduling time period, optimizes the energy interaction between the electric vehicle charging facilities in the building and the parking lot, further optimizes the load variance, and makes the operation of the building energy system more stable. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 It is a technical flow chart of the present invention;

[0087] Figure 2 This is the Monte Carlo extraction process under the full slow charging scenario;

[0088] Figure 3 This is the electric vehicle charging load distribution diagram under the full slow charging scenario;

[0089] Figure 4 To optimize the load variance of all possible situations by considering the uncertainty strategy;

[0090] Figure 5 is the total building power load under different strategies;

[0091] Figure 6 is the variance of the total building power load under different strategies;

[0092] Figure 7 This is the load comparison between Case a and Case-Ref. DETAILED DESCRIPTION

[0093] The present invention will be further described below by way of specific examples and accompanying drawings. The examples of the present invention are provided to help those skilled in the art better understand the present invention, and are not intended to limit the present invention in any way.

[0094] like Figure 1 As shown, this embodiment provides a building energy optimization scheduling method considering the uncertainty of electric vehicle charging load, including the following steps:

[0095] Step 1: Build a day-ahead forecast model for electric vehicle charging load

[0096] The parking lot of a small or medium-sized scientific research office building in Tianjin was selected as a typical case. The questionnaire survey was conducted by handing out questionnaires in person, and online questionnaires were filled out through two channels: forwarding by friends and family, and paid filling in on forums. A total of 500 questionnaires were distributed through different channels, and 427 valid questionnaires were collected. Six indicators were obtained, including the time of arrival at the workplace, the time of leaving the workplace, the driving distance from home to the workplace, the SOC at departure, the battery capacity, and the probability density distribution function of the power consumption per kilometer. All the charging piles in the parking lot are slow charging piles. According to the charging load calculation method introduced above, 10,000 Monte Carlo simulations were performed, such as Figure 2 As shown in the figure, the hourly distribution of electric vehicle charging load in this scenario is as follows: Figure 3 shown.

[0097] Step 2: Quantify charging load uncertainty

[0098] Based on the charging load prediction model, errors are introduced using the Monte Carlo method, and corrections are made based on the expected values ​​of each parameter. The next day's charging load samples, obtained after a sufficient number of Monte Carlo simulations, are combined to form a set of all possible charging scenarios for the next day. Real-time building load data is primarily obtained through the IoT energy consumption information monitoring platform. This platform monitors the building's real-time hourly cooling load and the power consumption of each device (building air conditioning load, as well as basic power loads such as lighting, office equipment, and elevators). Data collected over two months of summer is combined with charging load data to form a set of all possible power load scenarios for the next day.

[0099] Step 3: Establish a building energy optimization scheduling model considering charging load uncertainty

[0100] Various scheduling strategies are used to optimize the operation of the building energy system. For this case, the building is set to no optimization, only charging load optimization scheduling, and joint optimization scheduling of air conditioning load and charging load, as shown in Table 1.

[0101] Table 1 Comprehensive optimization scheduling strategy table

[0102]

[0103]

[0104] (1) Arrival and full-load non-guarantee rates

[0105] First, we used Case A to compare the optimization effects of a deterministic optimization scheduling strategy with an optimization scheduling strategy that considers charging load uncertainty. To verify the adaptability of the strategy derived from the deterministic method to all possible scenarios for the next day, we applied it to 10,000 scenarios, calculated the load variance under the worst-case scenario, and calculated the arrival and full charge uncertainty rates. These results were then compared with those of the optimization scheduling strategy that considers charging load uncertainty. The specific calculation results are shown in Table 2.

[0106] Table 2 Comparison of deterministic and uncertain optimization scheduling strategies

[0107]

[0108] Table 2 shows that the deterministic strategy's arrival and charging failure rates are both greater than the specified upper limit of 10%. When using the deterministic strategy, the designated charging start time would fail for 12.23% of vehicles facing various scenarios that could arise the next day, and 24.03% of vehicles would not be able to meet charging requirements upon departure. However, the strategy that considers the uncertainty of EV charging load shows a significant decrease in both arrival and charging failure rates, remaining within 10%. This demonstrates that the strategy that considers uncertainty is far more adaptable to the various scenarios that could arise the next day than the deterministic strategy.

[0109] (2) Load variance optimization

[0110] The strategy considering the uncertainty of charging load improves the load variance of various possible situations as follows: Figure 4 As shown. The strategy that considers uncertainty can optimize all possible scenarios, among which the load variance of most scenarios can be reduced by 5% to 10%. This shows that the strategy that considers uncertainty proposed by the present invention can effectively optimize the load variance of any possible scenario while ensuring that the strategy can be effectively implemented for more than 90% of vehicles and that the charging needs of more than 90% of vehicles can be met. Compared with the deterministic strategy, the strategy that considers charging load uncertainty proposed by the present invention does have superior robustness.

[0111] ① EV orderly charging

[0112] The total building power load curves for each strategy are as follows: Figure 5 As shown in the figure, the load variance of each strategy is as follows Figure 6 First, according to Figure 5 and Figure 6From Case a and Case-Ref, we can see that, compared with not adopting a scheduling strategy, orderly charging of electric vehicles can reduce the load variance by 2.83%, significantly reducing the load at 9:00 and 10:00 during the peak hours, and alleviating the negative effect of "peak on peak" after electric vehicles are connected. The main reason is that the charging start time of electric vehicles is re-determined, which realizes the transfer of charging load. The specific charging load is as follows: Figure 7 As shown in the figure, without orderly charging, EV loads are concentrated between 8:00 AM and 2:00 PM, particularly between 9:00 AM and 11:00 AM, coinciding with peak hours. However, with orderly charging, the EV charging load is evenly distributed between 9:00 AM and 7:00 PM, shifting the morning peak load to the parity period of 2:00 PM and 6:00 PM. This peak load shifting not only smoothes the load flow but also improves system stability.

[0113] from Figure 7 It can also be seen that due to the influence of car owners' commuting behavior, the time period in which electric vehicles can be adjusted is limited to between 8:00 and 19:00. Although the peak load can be transferred to the flat section, the valley section is still unutilized, and there is still great potential for optimizing the system load variance.

[0114] ② EV orderly charging + early start-up and pre-cooling

[0115] Further comparison Figure 5 and Figure 6 In Cases a and b, we can see that air conditioning precooling shifts the peak load at 8:00 AM to the nighttime valley period, significantly reducing the sudden increase in air conditioning load during morning startup. The load variance decreases by 3.06% to 17.81% compared to Case-Ref, achieving better optimization results than Case a, which only involves orderly EV charging. This demonstrates that air conditioning loads and EV charging loads complement each other well in terms of scheduling. Combining them for demand-side load management not only optimizes load variance and stabilizes the building energy system, but also shifts more load to valley periods, achieving greater peak-shaving and valley-filling.

[0116] In addition, compared Figure 6 From Case b-2h to Case b-6h, we can see that the earlier the pre-cooling starts, the smaller the load variance of the strategy and the better the system stability. This is because the earlier the pre-cooling starts, the longer the air conditioning runs at night, the greater the nighttime air conditioning load, which reduces the difference between daytime and nighttime air conditioning loads and reduces the load variance.

[0117] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A building energy optimization scheduling method considering the uncertainty of electric vehicle charging load, characterized by: The following steps are involved: Step 1): Establish a day-ahead forecast model for electric vehicle charging load Conduct a questionnaire survey in a specific region to obtain actual electric vehicle load data. Based on the collected data, analyze user commuting behavior characteristics and the distribution of electric vehicle physical properties, and establish an electric vehicle charging load prediction model based on Monte Carlo simulation and statistical methods. Step 2): Quantify charging load uncertainty Based on the charging load prediction model, errors are introduced through the Monte Carlo method, and corrections are made based on the expected values ​​of various parameters. The next day's charging load samples, obtained after a sufficient number of Monte Carlo simulations, are combined into a set to form a collection of all possible charging scenarios for the next day. The building's air conditioning load under pre-cooling conditions, as well as basic power loads such as lighting, office equipment, and elevators in the building, are measured and combined with the charging load to obtain a collection of all possible power load scenarios for the next day. Step 3): Establish a building energy optimization scheduling model considering charging load uncertainty A single-objective genetic algorithm is used to optimize the most unfavorable load variance in all possible scenarios. The charging start time of each electric vehicle is used as a variable, and the arrival and charging non-guaranteed rates are proposed as constraints to establish an optimal scheduling model for electric vehicle charging facilities. Specifically, (1) Determine the optimization variables During the operation phase, the sources of uncertainty in the electric vehicle charging load are the arrival time and the SOC at the time of arrival. To ensure the stable operation of the building energy system, uncertainty is introduced into the electric vehicle charging load. The start charging time of each electric vehicle on the second day is used as the optimization variable. Based on the prediction results of the second day's charging load, the start charging time of each electric vehicle on the second day is determined. (2) Determine the objective function Since excessive load fluctuations will seriously affect the operational stability and safety of the power grid, the load variance of the total power load is used as the optimization target. Considering the uncertainty, various charging load curves may appear on the second day. Therefore, the final strategy obtained by optimization should optimize all possible scenarios on the second day as much as possible. That is, no matter what Ω appears, EV In either case, the load variance should be improved after using the optimization strategy. The objective function is determined to minimize the maximum load variance. The calculation formula is as follows: Where Object is the objective function, which means minimizing the maximum value (the most unfavorable case) of the load variance in all scenarios; Net power load of group i at time t, kW; is the average value of the total power load of group i, kW; (3) Determine the constraints ① Electric vehicle starting charging time constraints The earliest charging start time for a single electric vehicle is the time of arrival. It is assumed that the SOC ≥ 0.8 when the vehicle leaves the vehicle to meet the owner's charging needs. Therefore, the latest charging start time plus the time required to charge to SOC = 0.8 cannot be later than the departure time of the vehicle. Where, is the minimum value among the 10,000 arrival times of the nth vehicle; t res,n The optimized charging start time for the nth vehicle; The time required for the nth vehicle in group i to charge to SOC = 0.8; The maximum value among the latest charging start times of the nth vehicle; ② Arrival rate is not guaranteed Since the arrival times of various vehicles on the next day vary, if the calculated charging start time for a vehicle on the next day is earlier than the vehicle's actual arrival time, the obtained strategy is considered invalid for that vehicle. It is necessary to wait until the vehicle arrives before re-assigning a charging start time for it, which reduces the effectiveness of the day-ahead optimization scheduling. To ensure the robustness of the obtained strategy and adapt to various situations, the concept of arrival non-guarantee rate is proposed: Where, N is the non-guaranteed rate at the destination; notarrive is the total number of vehicles whose arrival time is later than the re-set charging start time in all scenarios; The stipulation is that the arrival failure rate must not exceed 10%, meaning that the strategy is effective for 90% of vehicles in all scenarios. Since this arrival failure is very easy to occur, directly using hard constraints within a certain range may cause the optimization calculation to fail. Therefore, soft constraints are set to penalize the objective function when the conditions are not met: Where ε1 is the slack variable, ≥0; ③Charging rate constraint is not guaranteed If a vehicle charges according to the policy's reset charging start time but does not reach SOC = 0.8 when leaving, the policy is considered to not meet the vehicle owner's charging needs and this situation should be avoided. To ensure the robustness of the obtained policy and that it can meet the user's charging needs as much as possible in various scenarios, the concept of charging non-guarantee rate is proposed: Where, The charging rate is not guaranteed; N notfull is the total number of vehicles whose SOC still does not reach 0.8 when leaving in all scenarios; Just like the arrival guarantee, the objective function is penalized when the conditions are not met: Where ε2 is the slack variable, ≥0.

2. The optimization scheduling method according to claim 1, characterized in that: The step 1) establishes a day-ahead prediction model for electric vehicle charging load, specifically: (1) Obtain the distribution of relevant parameters through questionnaire survey To understand employees' commuting patterns, the questionnaire included basic employee information, travel habits, and physical parameters of electric vehicles. Analysis of the questionnaire results revealed six indicators: arrival time at work, departure time, distance traveled from home to work, SOC at departure, battery capacity, and the probability density distribution function of power consumption per kilometer. (2) Calculation of hourly charging load for a single vehicle To simplify the calculation of charging load, some assumptions and limitations are made to the conditions to a certain extent, as follows: ① Electric vehicles will not leave on the way to work and will be parked until the end of the working day; ② The electric vehicle will start charging when it arrives at the station and will continue charging until the battery is fully charged. If it cannot be fully charged by the time of departure, it will continue charging until the time of departure; ③ Since the duration of the charging start and end is very short, the charging process is regarded as constant power, and the slow charging power in this article is 7kW according to the national standard; ④Charging pile type configuration only establishes full slow charging scenario After obtaining samples of each random variable through Monte Carlo extraction, the SOC of each electric vehicle when it arrives at the station can be calculated: Where, SOC arr,n SOC is the state of charge when the nth vehicle arrives at the work site; de,n K is the state of charge of the nth car when it leaves home; n is the power consumption per kilometer of the nth vehicle, kWh / km; D n is the driving distance of the nth vehicle from home to work, km; Cap n is the battery capacity of the nth vehicle, kWh; From this, the charging time for each vehicle can be calculated: Where, L n is the charging time of the nth vehicle, hours; P c,n is the charging power of the nth vehicle, kW. In the full slow charging scenario, the charging power of all vehicles is 7kW; Assume that the charging state variable ε of the nth electric vehicle at a certain time t in one day is n (t): Where, t s,n is the moment when the nth vehicle starts charging, and the vehicle charging time ε n (t)=1, when not charging, it is ε n (t) = 0; The hourly charging load of a single vehicle and multiple vehicles is: P EV,n (t)=P c,n ×ε n (t) (10) Where, P EV,n (t) is the charging load of the nth vehicle at time t, kW; P EV (t) is the charging load of all vehicles at time t, kW; N is the number of all vehicles, and the hourly charging load of electric vehicles of a certain scale in a day can be obtained; (3) Monte Carlo simulation process For the full slow charging scenario, starting from the first vehicle, the start time of charging, electric vehicle battery capacity, power consumption per kilometer, distance from home to work, and SOC at departure are randomly sampled according to the probability density distribution function of each random variable. The hourly charging load of the vehicle is calculated according to formulas (7) to (11). The same sampling and calculation are performed for the remaining vehicles. The hourly charging load is accumulated to complete one Monte Carlo simulation process. The above sampling process is repeated to obtain a sufficient number of samples. This paper performs a total of 10,000 samplings. After the sampling is completed, 10,000 groups of hourly electric vehicle charging load sample data can be obtained. Then, the maximum, minimum, and median of the electric vehicle charging load at each moment are statistically analyzed and calculated to obtain the distribution characteristics of the electric vehicle load.

3. The optimization scheduling method according to claim 1, characterized in that: The step 2) quantifies the uncertainty of the charging load, specifically: (1) Correction of expectations and extraction of errors For each vehicle, in order to simulate the uncertain charging load prediction, the possible errors of the uncertain factors are extracted, and the expected values ​​of each factor are corrected to form a set of "situations to be considered"; The arrival time expression considering uncertainty deviation is: Where, t s,n (exp) is the expected value of the time when the nth vehicle arrives, i.e., the time when charging starts; For the predicted error value of the arrival time, a finite but sufficiently large set of error vectors is generated using the Monte Carlo method: Where, is the error vector of the arrival time prediction value, which contains all possible prediction errors; is the error observation vector extracted by Monte Carlo simulation; is the predicted value of the arrival time of the i-th extraction; Ne is the number of Monte Carlo extractions. When Ne is large enough, it can be considered that yes A fairly good approximation; let Ne = 10000 and the prediction error follow a normal distribution: Where, is the standard deviation of the arrival time prediction error; is the expected error in arrival time prediction; Similarly, the expression of the departure time considering uncertainty deviation is: Where, t l，n (exp) is the expected value of the latest time when the nth vehicle leaves, i.e., the end of charging; is the prediction error value of the departure time; is the error vector of the predicted value at the departure time, which contains all possible prediction errors; is the error observation vector extracted by Monte Carlo simulation; is the predicted value of the departure time extracted for the i-th time; is the probability density distribution function of the departure time prediction error; is the standard deviation of the prediction error of the departure time; is the expectation of the prediction error of the departure time; Similarly, the expression of SOC when considering uncertainty is: Where, SOC arr,n (exp) is the expected value of SOC when the nth vehicle arrives; is the predicted error value of SOC at arrival; is the error vector of the predicted SOC value at arrival, which includes all possible prediction errors; is the error observation vector extracted by Monte Carlo simulation; is the predicted SOC value at arrival extracted for the i-th time; is the probability density distribution function of the SOC prediction error at arrival; is the standard deviation of the SOC prediction error at arrival; is the expected SOC prediction error at arrival; (2) Forming a set of all possible charging loads After obtaining the calculation expression that takes into account the uncertainty of each variable, the Monte Carlo extraction method is used to extract the parameters of all vehicles each time. The results of each extraction are used as a set of possible situations for the next day. The extraction is repeated 10,000 times. By calculating the charging load, 10,000 sets of hourly charging load prediction results can be obtained. This set is used as all possible charging load curves for the next day, as shown in the following formula: Where, Ω EV is the charging load vector, which includes all possible situations on the next day; is the charging load observation vector extracted by Monte Carlo simulation; is the charging load of electric vehicles in group i within one day; is the charging load of electric vehicles in group i at time t; (3) Forming a set of all possible total power loads P s (t)=A×I(t)×η d (24) Where, Net power load of group i at time t, kW; P AC (t) is the air conditioning load at time t, kW; P other (t) is the other building basic power load at time t, including lighting, equipment, elevators, etc., in kW; is the average value of the total power load of group i, kW; P s (t) is the photovoltaic output at time t, which is 0 in the no-PV scenario, kW; A is the area of ​​the photovoltaic panel, m 2 ; I(t) is the total radiation intensity on the photovoltaic panel at time t, W / m 2 ; η d is the photovoltaic conversion efficiency.

Images