CN107239883A - A kind of dispatching method of Car sharing vehicle - Google Patents

Abstract

The invention relates to a scheduling method for vehicles in a car sharing system, comprising the following steps: 1) setting the failure probability of a site and obtaining a corresponding threshold according to the historical scheduling data of each site; 2) using the vehicle number and threshold of each site after the scheduling is completed The minimum difference is used as the objective function of the scheduling model, and set up constraints and priority settings, establish a scheduling model, and reset the threshold for events that need to be re-optimized; 3) Solve the scheduling model to obtain the corresponding scheduling strategy . Compared with the prior art, the present invention has the advantages of model assumptions, less parameters, simple calibration, scientific and effective, reasonable optimization model objectives, convenient realization, rapid solution, model constraint conditions in line with realistic requirements, and the like.

Description

A scheduling method for vehicles in a car sharing system

technical field

The invention relates to the field of vehicle allocation in a car sharing system, in particular to a scheduling method for vehicles in a car sharing system.

Background technique

The existing car-sharing system vehicle scheduling methods mainly include artificial experience judgment method, static linear programming method and dynamic stochastic programming method:

1. (Threshold-based) artificial experience judgment method

Based on experience, set the upper and lower thresholds for the number of vehicles at the site; compare the current number of vehicles with the threshold to obtain the scheduling requirements; and then employees can judge by themselves to obtain the scheduling plan.

Disadvantages: Threshold setting has no scientific basis; scheduling plan generation is not optimized.

2. (Threshold-based, minimum cost as the goal) static linear programming method

Set the upper and lower thresholds for the number of vehicles at the station; add the scheduling cost between stations; use linear programming to solve the vehicle deployment plan with the goal of minimizing the cost.

Disadvantages: There is no scientific basis for threshold setting; vehicle scheduling is a daily short-term decision; and factors related to scheduling costs, such as the number of employees, are fixed factors in the short term; therefore, cost is unreasonable as a target; static models cannot cope with dynamic changes in the system.

3. (Based on reliability) dynamic stochastic programming method

Factors such as user needs and usage time are used as random variables, and reliability indicators are introduced to dynamically optimize and solve the vehicle allocation plan.

Disadvantages: The parameter calibration in the use of the model is complex, and requires a large amount of data, which is prone to the problem of data sparseness; it is difficult to solve stochastic programming.

Contents of the invention

The purpose of the present invention is to overcome the above-mentioned defects in the prior art and provide a vehicle with model assumptions, few parameters, simple calibration, scientific and effective, reasonable optimization model objectives, convenient realization, rapid solution, and model constraints in line with actual requirements. A method for dispatching shared system vehicles.

A method for dispatching vehicles in a car sharing system, comprising the following steps:

1) Set the site failure probability and obtain the corresponding threshold according to the historical scheduling data of each site;

2) Take the minimum difference between the number of vehicles at each site and the threshold after the scheduling is completed as the objective function of the scheduling model, and set up constraints and priority settings, establish a scheduling model, and reset the threshold for events that need to be re-optimized ;

3) Solve the scheduling model to obtain the corresponding scheduling strategy.

In the described step 2), the objective function of the scheduling model is:

Among them, x _ij is the number of dispatched vehicles between the station that needs to call out the vehicle and the station j that needs to call in the vehicle within the optimization period T, α _i and α _j are the priority additional coefficients, and I _excess is the station where the vehicle is called out Set, I _shortage is the set of stations transferred into the vehicle.

In the described step 2), the priority setting includes site priority and state priority. In the described site priority, the priority of the site is represented by setting the priority additional coefficient. In the described state priority, set the site full load platform Takes precedence over site vacancy.

In the step 2), the constraint conditions include the upper limit constraint of the scheduling amount, the length constraint of the employee travel chain, the endurance constraint, the node conservation constraint, the single scheduling distance limit constraint and the feasibility constraint.

The upper limit of the dispatching quantity constraint is that the number of dispatching tasks should ensure that the number of dispatched vehicles at each site is not more than the actual dispatched vehicles, that is, not more than the dispatching demand, the expression is:

Among them, Veh _i is the current number of vehicles at station i, is the upper threshold of site i, is the lower threshold of site i.

The employee travel chain length constraint is:

in, is the scheduling task vector from site i to j, is the basis vector of Euclidean space R ⁿ , Dist _ij is the distance between station i and station j, V is the driving speed, TL _k is the actual limit value of the travel length of employee k, K is the employee number set, is the k-th dimension of the scheduling task vector from site i to j, that is, the number of times the k-th employee goes from site i to site j.

The endurance constraints described are:

in, with are the 1st and 2nd largest cruising ranges of vehicles at site i respectively, for The k _1th dimension of , for The k _2th dimension of , is the distance from site i to site j ₁ , are the distances between site i and site j ₂ respectively, and the first formula indicates that any employee starting from site i, the dispatching task distance to any site is less than the maximum cruising range of the vehicle at site i. The second formula shows that the sum of the scheduling tasks of any two employees starting from station i to any station (ie for any j ₁ ,j ₂ ∈I,k ₁ ,k ₂ ∈K) is less than the vehicle at station i The sum of the largest two values of the cruising range

The node conservation constraints described are:

in, In order to determine whether employee k is in the status of site i.

The single scheduling distance restriction constraint is:

Among them, Dist _max is the maximum distance limit for a single scheduling task.

The stated feasibility constraints are:

Among them, Fasb _ij (k) is the additional feasibility constraint on employee k from site i to j.

The purpose of the present invention can be achieved through the following technical solutions:

Compared with the prior art, the present invention has the following advantages:

1. Model assumptions, fewer parameters, and simple calibration

Only two assumptions (randomness of X _n and stability of parameter p) and three parameters (optimization period T, full-load control probability Prob _u and vacant control probability Prob _l ) are involved in this anti-wear; assumptions are made for the actual system The verification and parameter calibration are more convenient.

2. Scientific and effective

In the present invention, the concept of "threshold" is strictly defined, and a scientific calculation method is proposed, which eliminates the impreciseness of previous empirical methods.

3. The goal of optimization model is reasonable

The goal of the optimization solution in this aspect is to restore the sites in the network to normal as much as possible under the fixed long-term factors such as the number of employees; compared with other methods to optimize the system cost, this method is more reasonable.

4. Convenience and fast solution

The goal and constraint of the optimization model in the present invention adopt a linear form, which can be realized by convenient software, and the linear form model can be solved quickly, meeting the actual demand.

5. Model constraints meet realistic requirements

In the optimization model of the present invention, the influence of the characteristics of the electric vehicle on the generation of the dispatching plan is considered; in particular, the constraints of the length of the travel chain and the battery mileage are considered; for this reason, a special linear method is proposed, namely "vector optimization method" and "Linear Decomposition Method".

Description of drawings

Fig. 1 is a flow chart of the method of the present invention.

Fig. 2 is a data flow chart of the method of the present invention.

Fig. 3 is a schematic diagram of a scheduling scheme in an embodiment.

detailed description

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

Example

As shown in Figure 1, the flow process of the present invention is as follows:

1) Use a random process (Random Walk Model with Barriers) to simulate changes in site status, use historical order data for parameter calibration, and then calculate the probability of site failure;

2) Use the failure probability to define the threshold, and then obtain the scheduling requirements of the station vehicles (this method is called "failure probability control method" here);

3) Comparing the current state of the site with the threshold value, the scheduling requirements of each site are obtained;

4) Use the network flow model to optimize and solve the vehicle scheduling scheme and personnel task allocation scheme; the purpose is to ensure that as many stations in the network as possible work normally under limited resources;

5) The scheduling model runs dynamically. When some changes occur in the system state, the current scheduling task is terminated, and the above steps are re-optimized.

As shown in FIG. 2 , the data flow of the method of the present invention includes scheduling demand calculation, scheduling plan generation and parameter calibration.

1) Scheduling demand calculation

Upper (lower) threshold Thrd ^U (Thrd ^L ): defined as the upper limit (lower limit) of the number of vehicles that the site should keep, so that the probability of the site being fully loaded (vacant) is less than the given value Prob _u (Prob _l ); and this condition is satisfied The maximum (minimum) value of , see formulas (1) and (2). p _z and q _z represent the probability of full load and vacant failure of the station within the optimization period T, respectively. Prob _{u is} the control probability of the full load of the station; Prob _l is the control probability of the vacancy of the station, which is a set parameter.

Scheduling demand Need: It is the difference between the current number of vehicles Veh at the station and the threshold Thrd, see formula (3). When the number of vehicles at the station is higher than the upper threshold, Need is the number of vehicles at the station minus the upper threshold. At this time, Need is greater than zero, indicating that the vehicle needs to be called out; when the number of vehicles at the station is lower than the lower threshold, Need is the number of vehicles at the station minus the threshold. Remove the lower threshold, at this time Need is less than zero, indicating that the vehicle needs to be transferred.

Take a pick-up or return of a car at the station as an event. Let X _n denote the nth event of a site. When the nth event is returning the car, _Xn takes the value +1; when the nth event is borrowing the car, _Xn takes the value -1, see formula (5). Assume that the probability of X _n taking a value of +1 is p, and the probability of taking a value of -1 is q=1-p. p is a model parameter that needs to be calibrated using historical orders.

p{X _n ＝+1}＝p, p{X _n ＝-1}＝1-p (5)

Use a random walk model to calculate probabilities. Let N be the expected number of events in the optimization period T, that is, the number of steps (epochs) in the random wandering model. Here, v _z,n and u _z,n are used to denote the probability of full-load failure and vacant failure of the station at the nth step, respectively. The calculations of p _z , q _z and u _z,n are shown in formulas (6) to (7); by exchanging positions of p and q in formula (7), and then substituting (az) for z, formula (8) is obtained, Find the value of v _z,n .

2) Scheduling plan generation

The optimization variable of this algorithm, x _ij , represents the total number of trips of all dispatchers between stations i and j in the optimization period T; when i belongs to the station that needs to call out vehicles (i∈I _excess ), j needs to call When the station of the vehicle is (j∈I _shortage ), then x _ij represents the total number of dispatching behaviors between station i and j, that is, the total number of dispatched vehicles. In addition, x _ij represents the driving path of the dispatcher to complete the dispatch. Let Veh _i denote the number of vehicles at station i at the beginning of the optimization period.

The objective function is to meet the scheduling requirements of all stations as far as possible, that is, to minimize the difference between the number of vehicles and the threshold of each station after the scheduling is completed. When station i is the station that needs to call out the vehicle, Indicates the total number of vehicles transferred from i; Represents the difference between the number of vehicles at the site and the upper threshold after the task is completed. The index j takes the value and I _shortage (the number set of the station to be transferred to) when summed, because the scheduling model in this chapter uses the inventory-balancing scheduling strategy. Similar calculations are also performed for stations that need to transfer vehicles. The total difference between the number of vehicles at all stations and the threshold can be obtained by summing all stations, see formula (9).

Removing the constant term in (9) will not affect the optimization result, so a more concise form can be seen in formula (10). The impact of the threshold is reflected in the judgment of the need to call out ("excess") and need to call in ("shortage") sites, as well as the "upper limit constraint on scheduling volume" mentioned below.

For a practical system, generally speaking, the sum of the scheduling tasks of all sites is too large to complete; it is necessary to determine which scheduling tasks have priority; here are two priority settings:

i) Site priority

If some sites are considered more important and need to be guaranteed to be in a normal state, priority can be given, such as some important transportation hub sites or sites with important market significance. It is realized by adding coefficient α _i to the difference in the number of vehicles, see formulas (9) and (10).

ii) State priority

Users cannot return the car after the station is full, or the vehicle cannot be charged after returning the car, so it is generally considered that a full station is a more serious problem than an empty station. Try to give different priority to full load and empty failure control, here are two ways to achieve. One is to set the station full load control probability Probu _u lower than the station vacancy control probability Prob _l , that is, the probability of allowing the station full load is smaller. Alternatively, it may be required that personnel must be dispatched to solve the scheduling needs of fully loaded sites, see formula (11).

Six types of constraints are listed in the scenario generation model, see i)-iv).

i) Scheduling volume upper limit constraint

The number of scheduling tasks should ensure that the number of vehicles scheduled for each site is not more than the actual number of vehicles that need to be scheduled, that is, not more than the scheduling demand. As shown in formulas (12) and (13), the left side of the greater than or equal sign is the number of vehicles at the station after the scheduling is completed, and the right side is the threshold of the station.

ii) Constraints on employee travel chain length

Usually the dispatcher will start from a dispatch center, work a shift (such as 4 hours), and then return to the dispatch center to rest or change shifts. The time from the current moment to the end of this shift is the available time of dispatcher k, recorded as TL _k . In this algorithm, a method of constraining the length of dispatcher's travel chain is proposed, and the constraint is guaranteed to be in a linear form (herein, this method is called "vector optimization method").

Suppose there are K employees at the beginning of the optimization cycle. "Split" each x _ij variable into K variable value sums, using a K-dimensional vector express. The kth dimension of , namely Indicates the number of vehicles dispatched by the kth employee from station i to station j. Under this representation method, the "tracking" of the path of each dispatcher can be realized. Obviously by definition. The sum of the dimensions of is equal to That is formula (14). Represents the travel time from station i to station j. Indicates the time it takes for the kth employee to execute the scheduling task from station i to station j; if employee k does not pass through the ij section, that is The corresponding kth dimension component is zero. By summing i and j, we get the total path time of the kth employee in the network. The length needs to be less than or equal to the employee's available time TL _k , see formula (15). Considering that the number of employees in the real system is generally small, constraint (16) can be added so that the same employee will not perform tasks at a pair of sites more than once during the optimization period, and the model becomes a 0-1 planning model; can be solved more quickly. Of course, this constraint can also be omitted, and only require If it is an integer, it is a general integer programming model.

iii) Endurance constraints

For the electric vehicle system, it is necessary to add endurance constraints when generating the scheduling plan to ensure that the generated scheduling tasks are practical; to ensure that there will be no situation where the vehicle is dispatched halfway without power. This requires that the distance of the nth furthest scheduling task of site i is less than the nth largest cruising power of the vehicle at the site These conditions are non-linear. In this algorithm, a set of linear ones is used instead, which is called "linear decomposition method". The nth set of constraints requires that the sum of the n tasks with the largest site distance be less than the sum of the largest n battery life. It should be noted that these conditions are necessary but not sufficient; the main purpose of using this condition is to solve more quickly and basically meet the actual needs. The nth set of constraints actually contains constraints. To consider all cases, there will be 2 ⁿ conditions. In order to avoid an exponential increase in the number of conditions in this set of conditions, it can be simplified to only consider the first few sets of conditions. Generally speaking, there are not too many scheduling tasks between the same pair of sites, and only the first two or three sets of conditions can be considered; this simplification is in line with the actual situation. The case where only the first two sets of conditions are considered includes formulas (17) and (18). If a constant is added to the condition, it can also be required that the vehicle has a certain amount of power remaining after it is transferred to the target station; to ensure that the vehicle can be used immediately. What needs to be explained here is that, in order to simplify the consideration, the charging process of the vehicle in the optimization cycle is ignored in this model. The vehicle cruising range is taken as the value at the beginning of the optimization period and is kept as a constant throughout the optimization period.

iv) Node Conservation Constraints

This set of constraints is used to ensure that employees do not "disappear" from the network. Under the new vector representation method, the staff is required to maintain "conservation" at each site and each dimension, see formula (19); that is, i and k need to be taken through the number set. Indicates the number of visits to this site from other sites. Indicates the number of departures from this site to other sites.

v) Single dispatch distance limit

This set of constraints is used to avoid some overly distant schedules. Scheduling over long distances consumes a lot of time and manpower, and this kind of scheduling is often avoided in practice. This is a limitation often added by the system operator, see formula (20). Dist _max is a parameter set by the operator itself, that is, the upper limit.

vi) Feasibility constraints

This set of constraints is used to represent the additional constraints. In order to realize the "equilibrium vehicle number" scheduling strategy, such constraints need to be added, requiring that the Fasb _ij (k) between the OD pairs of other stations be zero except for the following station OD pairs, that is, the dispatcher is not allowed to pass.

Departure link (Staff departure link): connects the OD pair between the dispatch center where the dispatcher departs and the site that needs to be called out (the dispatcher goes to perform the first task);

Vehicle relocating links: connect the OD pair between the station that needs to be called out (i∈I _excess ) and the station that needs to be called in (j∈I _shortage ) (the dispatcher executes a task);

Non-scheduling links (Staff rebalancing links): connect the OD pair between the station that needs to be called in (i∈I _shortage ) and the station that needs to be called out (j∈I _excess ) (the dispatcher is rushing to the next station that needs to be called out section of the site);

Return links (Staff returning links): connect the station that needs to be called in and the dispatch center (the dispatcher returns to the dispatch center).

Integer programming model is used to solve the above network flow model. The trip chain of the generated scheduling scheme is shown in Figure 3.

In order to make the model respond to the real-time changes of the system state, a dynamic optimization rolling method is adopted. The events that need to trigger reoptimization are marked below. In addition to those listed below, if there are actually other reasons to re-optimize, the scheduling algorithm can also be immediately re-invoked.

i) The status of the site is abnormal, such as when the site is fully loaded or vacant;

ii) The set optimization period ends;

iii) The total state change of the system (as measured by the total change of vehicle inventory level at the station) reaches the specified value;

iv) Employee status changes (such as employee shift change, new employees).

When re-optimizing, there are two ways to calculate the threshold of the new optimization period: one is online (on-line) calculation, according to the re-optimization start time point, call historical order data, and use model calculation; without the support of high-performance computers This method takes a long time and is not recommended; another method is off-line calculation; at intervals of about 30 minutes, calculate the threshold corresponding to the T length optimization cycle starting at each time point.

The meanings of the symbols involved in the brief description of the above principles are shown in Table 1.

Table 1 Meaning table of symbols

3) Parameter calibration

The setting of the optimization period T should make the parameter p stable during this period, that is, the assumption of the scheduling demand generation model is established. The selection of the T length makes the multi-day sample data of the parameter p obey the normal distribution, so the sample mean can be used for estimation. For practical systems, T should generally not be set to less than 2 hours. This article recommends a T setting between 3 and 6 hours.

The lower the Prob _u and Prob _l are set, the lower the probability of site failure is allowed, and the higher the scheduling requirement is. If set too low, it will generate scheduling requirements that are practically impossible to fulfill. If the setting is too high, labor unions may cause waste of manpower. Through analysis, it is concluded that the reasonable value range of Prob _u and Prob _l is 0.3–0.7.

Claims (10)

1. A dispatching method of a car sharing system vehicle, characterized in that, comprising the following steps: 1) Set the site failure probability and obtain the corresponding threshold according to the historical scheduling data of each site; 2) Take the minimum difference between the number of vehicles at each site and the threshold after the scheduling is completed as the objective function of the scheduling model, and set up constraints and priority settings, establish a scheduling model, and reset the threshold for events that need to be re-optimized ; 3) Solve the scheduling model to obtain the corresponding scheduling strategy. 2. the scheduling method of a kind of car sharing system vehicle according to claim 1, is characterized in that, in described step 2), the objective function of scheduling model is: <mrow> <mi>min</mi> <mo>{</mo> <munder> <munder> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>&amp;Element;</mo> <msub> <mi>I</mi> <mrow> <mi>e</mi> <mi>x</mi> <mi>c</mi> <mi>e</mi> <mi>s</mi> <mi>s</mi> </mrow> </msub> </mrow> </munder> <mrow> <mi>j</mi> <mo>&amp;Element;</mo> <msub> <mi>I</mi> <mrow> <mi>s</mi> <mi>h</mi> <mi>o</mi> <mi>r</mi> <mi>t</mi> <mi>a</mi> <mi>g</mi> <mi>e</mi> </mrow> </msub> </mrow> </munder> <mrow> <mo>(</mo> <msub> <mi>&amp;alpha;</mi> <mi>i</mi> </msub> <mo>+</mo> <msub> <mi>&amp;alpha;</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <msub> <mi>x</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mo>}</mo> </mrow> Among them, x _ij is the number of dispatched vehicles between the station i that needs to call out the vehicle and the station j that needs to call in the vehicle within the optimization period T, α _i and α _j are the priority additional coefficients, and I _excess is the number of vehicles that are called out Station collection, I _shortage is the station collection transferred to the vehicle. 3. The dispatching method of a kind of car sharing system vehicle according to claim 2, it is characterized in that, in the described step 2), priority setting includes site priority and state priority, in the described site priority, by setting priority The additional coefficient of the weight is used to represent the priority of the station. In the state priority, the priority of the station that is fully loaded is higher than that of the vacant station. 4. The dispatching method of a kind of car sharing system vehicle according to claim 2, it is characterized in that, in described step 2), constraint condition comprises dispatching amount upper limit constraint, employee travel chain length constraint, battery life constraint, node conservation Constraints, single-scheduling distance limit constraints, and feasibility constraints. 5. A method for scheduling vehicles in a car sharing system according to claim 4, wherein the upper limit of the scheduling amount is constrained to ensure that the number of scheduled tasks at each site is no more than the actual number of vehicles required The dispatched vehicles, that is, not more than the dispatching demand, the expression is: <mrow> <msub> <mi>Veh</mi> <mi>i</mi> </msub> <mo>-</mo> <munder> <mo>&amp;Sigma;</mo> <mi>j</mi> </munder> <msub> <mi>x</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mo>&amp;GreaterEqual;</mo> <msubsup> <mi>Thrd</mi> <mi>i</mi> <mi>U</mi> </msubsup> <mo>,</mo> <mi>f</mi> <mi>o</mi> <mi>r</mi> <mi> </mi> <mi>e</mi> <mi>a</mi> <mi>c</mi> <mi>h</mi> <mi> </mi> <mi>i</mi> <mo>&amp;Element;</mo> <msub> <mi>I</mi> <mrow> <mi>e</mi> <mi>x</mi> <mi>c</mi> <mi>e</mi> <mi>s</mi> <mi>s</mi> </mrow> </msub> </mrow> <mrow> <msub> <mi>Veh</mi> <mi>i</mi> </msub> <mo>+</mo> <munder> <mo>&amp;Sigma;</mo> <mi>i</mi> </munder> <msub> <mi>x</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <mo>&amp;le;</mo> <msubsup> <mi>Thrd</mi> <mi>i</mi> <mi>L</mi> </msubsup> <mo>,</mo> <mi>f</mi> <mi>o</mi> <mi>r</mi> <mi> </mi> <mi>e</mi> <mi>a</mi> <mi>c</mi> <mi>h</mi> <mi> </mi> <mi>i</mi> <mo>&amp;Element;</mo> <msub> <mi>I</mi> <mrow> <mi>s</mi> <mi>h</mi> <mi>o</mi> <mi>r</mi> <mi>t</mi> <mi>a</mi> <mi>g</mi> <mi>e</mi> </mrow> </msub> </mrow> Among them, Veh _i is the current number of vehicles at station i, is the upper threshold of site i, is the lower threshold of site i. 6. The dispatching method of a kind of car sharing system vehicle according to claim 4, is characterized in that, described employee's travel chain length constraint is: in, is the scheduling task vector from site i to j, is the basis vector of Euclidean space R ⁿ , Dist _ij is the distance between station i and station j, V is the driving speed, TL _k is the actual limit value of the travel length of employee k, K is the employee number set, is the k-th dimension of the scheduling task vector from site i to j, that is, the number of times the k-th employee goes from site i to site j. 7. The scheduling method of a car sharing system vehicle according to claim 6, wherein the endurance constraint is as follows: in, with are the 1st and 2nd largest cruising ranges of vehicles at site i respectively, for The k _1th dimension of , for The k _2th dimension of , is the distance from site i to site j ₁ , are the distances from site i to site j ₂ respectively. 8. The dispatching method of a kind of car sharing system vehicle according to claim 6, is characterized in that, described node conservation constraint is: in, In order to determine whether employee k is in the status of site i. 9. The dispatching method of a kind of car sharing system vehicle according to claim 8, is characterized in that, described single dispatching distance limit constraint is: Among them, Dist _max is the maximum distance limit for a single scheduling task. 10. The scheduling method of a car sharing system vehicle according to claim 8, wherein the feasibility constraint is: Among them, Fasb _ij (k) is the additional feasibility constraint on employee k from site i to j.