An electric vehicle charging pile order exception clearing and settling method and device

Abstract

The application discloses a kind of electric vehicle charging pile order exception clearing settlement method and device, comprising: obtaining the real-time order of vehicle networking platform;Based on the inference engine of preset detection real-time order whether it is abnormal order;Response to real-time order is normal order, then directly upload real-time order to vehicle networking platform;Response to real-time order is abnormal order, then based on correction algorithm or preset prediction model carries out numerical correction or data prediction to real-time order, obtains revised abnormal order and uploads to vehicle networking platform.One aspect, according to the mass historical abnormal data obtained by vehicle networking platform, mining clearing screening rule, combined with the experience rule of power grid expert custom clearing screening rule, the inference engine obtained can accurately detect abnormal order, on the other hand, it can be to electric vehicle charging pile abnormal order numerical correction or data prediction, reduce the economic loss caused by order exception.

Description

Technical Field

[0001] This invention relates to a method and apparatus for clearing and settling abnormal orders for electric vehicle charging stations, belonging to the field of electric vehicle charging station operation and management technology. Background Technology

[0002] With the continuous development of the new energy industry, the number of new energy electric vehicles in major cities across the country is increasing, leading to a rise in the utilization rate of charging piles. This has also resulted in a growing volume of related data, such as electronic orders generated during the charging process. Due to the independent management of vehicle-to-everything (V2X) platforms and the complexity of their deployment environments, the possibility of abnormal or lost transaction records is increasing, causing problems in areas such as charging amount verification and settlement.

[0003] The main applications of data clearing algorithms are in urban rail transit and ticketing clearing. However, there is currently a lack of research in China on algorithms for charging facility order lists. Accurately predicting the time periods of abnormal data during electric vehicle charging is key to the research of clearing algorithms, and this will be the main research direction for electric vehicle charging pile data clearing in the future.

[0004] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and device for clearing and settling abnormal electric vehicle charging pile orders. On the one hand, based on the massive historical abnormal data obtained from the vehicle network platform, clearing and screening rules are mined and customized by combining the experience rules of power grid experts. The resulting inference engine can accurately detect abnormal orders. On the other hand, it can correct the values ​​or predict the data of abnormal electric vehicle charging pile orders, thereby reducing the economic losses caused by abnormal orders.

[0006] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0007] On one hand, this invention discloses a method for clearing and settling abnormal electric vehicle charging station orders, including:

[0008] Obtain real-time orders from the vehicle networking platform;

[0009] The system uses a pre-set inference engine to detect whether the real-time order is an abnormal order.

[0010] If the real-time order is a normal order, then the real-time order is directly uploaded to the vehicle networking platform;

[0011] In response to the real-time order being an abnormal order, the real-time order is numerically corrected or data is predicted based on a correction algorithm or a preset prediction model to obtain a revised abnormal order and upload it to the vehicle network platform.

[0012] The method for constructing the inference engine includes:

[0013] Obtain historical abnormal orders from the vehicle networking platform;

[0014] Based on the aforementioned historical abnormal orders, and using an improved decision tree mining algorithm, the sorting and screening rules are obtained.

[0015] Based on the mining and sorting rules, combined with custom sorting and sorting rules, an inference engine is obtained.

[0016] Furthermore, the abnormal orders include orders with abnormal data and orders with missing data.

[0017] In response to the fact that the abnormal order is a data abnormal order, the data abnormal order is numerically corrected based on the correction algorithm to obtain the revised data abnormal order and uploaded to the vehicle network platform;

[0018] In response to the fact that the abnormal order is a missing abnormal order, the missing abnormal order is numerically corrected based on a preset prediction model to obtain a revised missing abnormal order and uploaded to the vehicle network platform.

[0019] Furthermore, the expression for the correction algorithm is as follows:

[0020]

[0021]

[0022] In the formula, v i ' _abnormal These are the outlier values ​​in orders with data errors, i.e., the order values ​​that need to be corrected. These are normal values ​​in the data anomaly bill; To and The corresponding normal value in a normal bill; α is the proportional coefficient; v i_normal For with v′ i_abnormal The corresponding normal value in a normal bill.

[0023] Furthermore, the method for constructing the prediction model includes:

[0024] Obtain historical normal orders from the vehicle networking platform and convert them into triplet form;

[0025] Based on the pre-defined permutation language model XLNet, the historical normal orders are split into triples to obtain the input sequence;

[0026] Based on the input sequence, a prediction model of a recurrent neural network is obtained by training a long short-term memory network.

[0027] Furthermore, the expression for the prediction model is as follows:

[0028] h′ t =f(W1x t +W3h′ t +b′ t )

[0029] h t =f(W2x) t +W4h t-1 +b t )

[0030]

[0031] y ＜T＞ =g(W hy ×h t +b y )

[0032] In the formula, x t This represents the input, specifically the numerical values ​​that are not missing from the core attributes; y <t>< / t> represents the output, i.e., the actual predicted missing values; f represents the activation function of the hidden layer of the recurrent neural network; W1, W2, W3, and W4 represent the first, second, third, and fourth power grid attribute association weight matrices after training through the recurrent neural network, respectively; h′ t Indicates the updated candidate state; h t-1 This represents the temporary value stored in the recurrent neural network for calculating the bill attributes at time t-1; h t This represents the temporary value calculated by the recurrent neural network for the billing attributes at time t; b′ t This represents the bias vector updated by the recurrent neural network based on the bill attributes; b t H represents the bias vector fitted by the recurrent neural network based on bill attributes; t W represents the input state sequence; hy The output layer weight matrix is ​​represented by g; the activation function is represented by b. y This represents the bias matrix of the output layer; This is a vector concatenation operation.

[0033] Furthermore, the expression for the improved decision tree mining algorithm is as follows:

[0034]

[0035]

[0036] In the formula, I represents the number of sorting and filtering rules selected by the inference engine; R i The number of abnormal orders that meet the sorting and filtering rules I; D is the total number of abnormal orders; p i Gini(p) represents the support level of orders that meet the i-th sorting and filtering rule in the vehicle network data platform. i ) represents the Gini coefficient of each rule, and the rule with the smallest Gini coefficient among the selected rules is taken as the root node of the decision tree.

[0037] Furthermore, the custom sorting and filtering rules include:

[0038] The order start time is later than the end time.

[0039] The transaction order volume is greater than 400;

[0040] The initial value of the electricity meter in the transaction order is greater than the meter reading.

[0041] The sum of the time-of-use electricity charges for each transaction order does not equal the total electricity charge.

[0042] The sum of the electricity consumption of each transaction at different times does not equal the total electricity consumption.

[0043] The sum of the time-based service fees for each transaction order does not equal the total service fee;

[0044] The starting value of this transaction order is greater than the ending value of the previous order by more than 0.5;

[0045] The number of times the charging completion status changed is inconsistent with the number of received transaction orders;

[0046] There is no corresponding transaction order when the charging completion status change time value is compared with the charging order completion time value.

[0047] Furthermore, obtaining real-time orders from the connected vehicle platform includes:

[0048] Obtain real-time order data from the vehicle networking platform.

[0049] After unifying the semantic attributes of the real-time order data based on the knowledge fusion algorithm, the data is transformed into a triplet form to obtain the real-time orders of the vehicle network platform.

[0050] The expression in the form of the triple is as follows:

[0051] {e i_NO. ,r i ,v i}

[0052] Among them, e i_NO. Represents the i-th order header entity, r i v represents the physical quantity of the i-th type of order in the order.i This represents the tail entity of the i-th order.

[0053] On the other hand, this invention discloses a clearing and settlement device for abnormal electric vehicle charging pile orders, used to implement the above-mentioned clearing and settlement method for abnormal electric vehicle charging pile orders, including:

[0054] Data acquisition module: used to acquire real-time orders from the vehicle networking platform;

[0055] Order detection module: used to detect whether the real-time order is an abnormal order based on a preset inference engine, and to obtain normal orders and abnormal orders;

[0056] Abnormal Order Revision Module: Used to perform numerical correction or data prediction on the abnormal orders based on correction algorithms or preset prediction models to obtain revised abnormal orders;

[0057] Data upload module: used to upload the normal orders and the revised abnormal orders to the vehicle networking platform.

[0058] Furthermore, the abnormal order revision module includes a data abnormal order revision module and a missing abnormal order revision module.

[0059] The data anomaly order revision module is used to perform numerical correction on data anomaly orders based on a correction algorithm to obtain revised data anomaly orders;

[0060] The missing and abnormal order revision module is used to perform numerical correction on missing and abnormal orders based on a preset prediction model to obtain revised missing and abnormal orders.

[0061] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0062] On the one hand, this invention mines clearing and screening rules based on massive historical abnormal data obtained from the vehicle network platform, and customizes clearing and screening rules by combining the experience rules of power grid experts. The resulting inference engine can accurately detect abnormal orders. On the other hand, it can perform numerical correction or data prediction for abnormal orders of electric vehicle charging piles, thereby reducing economic losses caused by abnormal orders.

[0063] This invention also acquires historical normal order data and trains it using a long short-term memory network to obtain a recurrent neural network prediction model. This prediction model can accurately identify missing values ​​in abnormal order data, thereby effectively addressing the economic losses caused by abnormal or lost order data and providing theoretical support for the vigorous development of my country's electric vehicle industry. Attached Figure Description

[0064] Figure 1This is a flowchart of a method for clearing and settling abnormal orders for electric vehicle charging stations;

[0065] Figure 2 This is a flowchart based on the nine rules for detecting abnormal bills;

[0066] Figure 3 It is a diagram of the real-time order reasoning process based on the reasoning engine;

[0067] Figure 4 It is a flowchart of rule mining based on decision trees;

[0068] Figure 5 It is a fitting curve based on a recurrent neural network training model. Detailed Implementation

[0069] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0070] Example 1

[0071] This embodiment 1 discloses a method for clearing and settling abnormal electric vehicle charging station orders, such as... Figure 1 As shown, it includes:

[0072] Obtain real-time orders from the vehicle networking platform;

[0073] The system uses a pre-set inference engine to detect whether real-time orders are abnormal.

[0074] If the real-time order is a normal order, then the real-time order will be directly uploaded to the vehicle networking platform;

[0075] If a real-time order is found to be an abnormal order, the real-time order will be numerically corrected or data predicted based on a correction algorithm or a preset prediction model to obtain a revised abnormal order and upload it to the vehicle network platform.

[0076] The methods for building an inference engine include:

[0077] Obtain historical abnormal orders from the vehicle networking platform;

[0078] Based on historical abnormal orders, and using an improved decision tree mining algorithm, sorting and screening rules are obtained.

[0079] Based on the mining and sorting rules, combined with custom sorting and sorting rules, an inference engine is obtained.

[0080] The technical concept of this invention is as follows: On the one hand, based on the massive historical abnormal data obtained from the vehicle network platform, sorting and screening rules are mined, and sorting and screening rules are customized by combining the experience rules of power grid experts. The resulting inference engine can accurately detect abnormal orders. On the other hand, it can correct the numerical values ​​or predict the data of abnormal orders for electric vehicle charging piles, thereby reducing the economic losses caused by abnormal orders.

[0081] Specific steps are as follows Figure 1 As shown:

[0082] Step 1: Data Acquisition.

[0083] We acquire massive amounts of data from the vehicle networking platform, including historical order data and real-time order data.

[0084] It should be noted that historical order data is used to build the inference engine and prediction model, and is not obtained at the same time as real-time order data, but both need to be processed as follows.

[0085] Knowledge fusion algorithms are used to effectively integrate the massive amounts of data acquired. For example, if different databases contain semantically identical attributes, such as "total charging amount" and "cumulative charging amount," a mapping process is required.

[0086]

[0087] In the formula, M is the mapping function; This refers to a specific attribute in the Internet of Vehicles (IoV). This is another attribute in the Internet of Vehicles; ψ is the definition method specified in the clearing algorithm.

[0088] Secondly, it is transformed into the standard triplet form {e} applicable to inference engines. i_NO .,r i ,v i}, e i Defined as the i-th order header entity, r i Let v be the physical quantity corresponding to the i-th type of order in the order. i Define the i-th order tail entity; obtain historical orders and real-time orders.

[0089] The following is an example of real-time order data:

[0090] Table 1 Charging data for a certain region in March

[0091]

[0092] Convert the above data into triples required by the inference engine:

[0093] (Transaction order 247, order status: "Paid")

[0094] (Transaction order 247, transaction volume 450)

[0095] (Transaction order 247, transaction amount: 119.54)

[0096] (Transaction order 247, creation time: 2022-03-29 19:06:59)

[0097] (Transaction order 247, start time: 2022-03-29 19:07:01)

[0098] (Transaction order 247, end time: 2022-03-29 21:23:08)

[0099] (Transaction order 247, payment time: 2022-03-29 21:33:18)

[0100] (Transaction order 369, order status: "Pending payment")

[0101] (Transaction order 369, transaction volume 20.07)

[0102] (Transaction order 369, transaction amount: 0)

[0103] (Transaction order 369, creation time: 2022-04-02 09:49:58)

[0104] (Transaction order 369, start time: 0)

[0105] (Transaction order 369, end time: 0)

[0106] (Transaction order 369, payment time: 0)

[0107] It should be noted that both transaction orders 247 and 369 are abnormal. Transaction order 247 has a transaction volume greater than 400, which will be further explained in step three; transaction order 369 is missing data, lacking transaction amount, start time, end time, and payment period. The purpose of transaction orders 247 and 369 here is only to demonstrate how to transform massive amounts of data into a standard triplet form applicable to the inference engine, and no other processing is performed.

[0108] Step Two: Order Inspection

[0109] 1) Before performing real-time order detection, build an inference engine to detect whether a real-time order is an abnormal order. The specific steps are as follows:

[0110] First, based on the experience and rules of power grid experts, custom sorting and screening rules are defined, including:

[0111] (1) The start time of the transaction order is later than the end time;

[0112] (2) The transaction order volume is greater than 400;

[0113] (3) The initial value of the electricity meter in the transaction order is greater than the value indicated by the electricity meter;

[0114] (4) The sum of the time-of-use electricity charges for each transaction order does not equal the total electricity charge;

[0115] (5) The sum of the time-of-use electricity consumption in each transaction order does not equal the total electricity consumption;

[0116] (6) The sum of the time-based service fees for each transaction order does not equal the total service fee;

[0117] (7) The starting value of this transaction order is greater than the ending value of the previous order by more than 0.5;

[0118] (8) The number of times the charging end status changed is inconsistent with the number of received transaction orders;

[0119] (9) There is no corresponding transaction order when the time value of the charging end status change is compared with the time value of the charging order ending charging.

[0120] Secondly, based on nine custom clearing and filtering rules, a statistical process for clearing data on the vehicle networking platform is constructed, as follows: Figure 2 As shown, the custom clearing and filtering rules are formally defined using the Datalog language and input into the high-performance inference engine. The nine custom clearing and filtering rules are as follows:

[0121] (1) Datalog rule 1: (Order_id, exception type, transaction time exception): -(Order_id, order status, "paid") ∧(Order_id, start time, T start )∧(T start Later than, T end ), where Order_id represents the order number, T start T end These represent the start and end times of the order, respectively.

[0122] (2) Datalog Rule 2: (Order_id, exception category, transaction data exception): -(Order_id, order status, "paid") ∧(Order_id, transaction amount, G) 交易 )∧(G 交易 (greater than 400), G 交易 Indicates the amount of electricity traded;

[0123] (3) Datalog rule 3: (Order_id, exception category, transaction electricity exception): -(Order_id, order status, "unpaid") ∧(Order_id, total meter reading, G start )∧(G start Greater than, G end ), G start G end These correspond to the total starting value and total ending value of the electricity meter, respectively.

[0124] (4) Datalog Rule 4: (Order_id, exception category, transaction data exception): -(Order_id, order status, "paid") ∧(Order_id, time-of-use electricity fee, Q 分 )∧(Q 分 , not equal to, Q 总 ), Q 分 Q 总 These represent time-of-use electricity charges and total electricity charges, respectively.

[0125] (5) Datalog Rule 5: (Order_id, exception category, transaction data exception): -(Order_id, order status, "paid") ∧(Order_id, time-of-use electricity, G 分 )∧(G 分 , not equal to, G 总 )G 分 G 总 These represent time-of-use electricity consumption and total electricity consumption, respectively.

[0126] (6) Datalog Rule 6: (Order_id, exception category, transaction data exception): -(Order_id, order status, "paid") ∧(Order_id, time-sharing service fee, S 分 )∧(S 分 , not equal to, S 总 ), S 分 S 总 These represent the time-based service fee and the total service fee, respectively.

[0127] (7) Datalog Rule 7: (Order_id, exception category, transaction data exception): -(Order_id, order status, "unpaid") ∧(Order_id, transaction amount, Q) i_start -Q i-1_end )∧(Q i_start -Q i-1_end (greater than, 0.5), Q i_start Indicates the starting value of the order's battery level, Q. i-1_end This indicates the termination value of the battery level for the last order;

[0128] (8) Datalog rule 8: (Order_id, exception category, transaction data exception): -(Order_id, order status, "unpaid") ∧(Order_id, number of order ends, Order_end) ∧(Order_end, not equal to, Order_reception), where Order_end represents the number of order ends and Order_reception represents the number of order received;

[0129] (9) Datalog Rule 9: (Order_id, exception type, missing transaction amount): -(Order_id, order status, "pending payment") ∧(Order_id, total electricity consumption, G 总 )∧(G 总 , greater than, 0.1 / kwh)∧(Order_id, transaction amount, S)∧(S, equal to, 0), where S represents the transaction amount.

[0130] Based on nine custom clearing and filtering rules, abnormal orders are filtered out from historical orders. These abnormal orders are then used as input for incremental mining based on knowledge graph technology. An improved decision tree mining algorithm is then applied. Figure 4 As shown, new detection rules are discovered, and the improved decision tree mining algorithm is as follows:

[0131]

[0132]

[0133] In the formula, I represents the number of sorting and filtering rules arbitrarily selected by the inference engine; R i The number of abnormal orders that meet the sorting and filtering rules I; D is the total number of abnormal orders; p i Gini(p) represents the support level of orders that meet the i-th sorting and filtering rule in the vehicle network data platform. i The Gini coefficient of each rule is represented by ), and the rule with the smallest Gini coefficient among the selected rules is taken as the root node of the decision tree; the weighted Gini coefficient Gini'(p) after rule mining. I+1 The formula is as follows:

[0134]

[0135] in This represents a weighted coefficient vector composed of the correlation degrees of each rule. This represents the Gini coefficient vector composed of the Gini coefficients of each rule. When the weighted Gini coefficients are calculated, Gini'(p) I+1 When the coefficient is less than the Gini coefficient of each of the selected rules, it indicates that a new rule has been generated by association with the selected rules.

[0136] Based on the new mining detection rules, combined with the custom sorting and filtering rules, the rules are input into the inference engine in the form of triples, and participate in the inference as the rule part of the inference engine, thus obtaining the inference engine used to detect abnormal orders in this embodiment.

[0137] 2) Use the pre-built inference engine for inference, and perform customized development using the high-performance inference engine Drools to complete the numerical inference detection of real-time equipment fault conditions, and detect whether real-time orders are abnormal orders, obtaining normal orders and abnormal orders. Normal orders proceed to step four, while abnormal orders, including data-abnormal orders and missing order abnormal orders, proceed to step three.

[0138] Simultaneously, based on fact triples and the Leaps algorithm in Drools' built-in rule engine, new correct triples containing abnormal order information are derived. For example, abnormal orders may include orders with missing data. The triple data is composed of text, but for orders with missing data, we need to complete them according to their missing attributes. For example, sometimes the battery attribute is missing, so the battery value is filled to 0 according to the fact triple. The inference process of real-time orders based on the inference engine is shown in the diagram below. Figure 3 As shown.

[0139] Step 3: Revise Abnormal Orders

[0140] 1) Orders with abnormal data

[0141] Orders with abnormal data are corrected according to prior rules to obtain orders with correct attribute values, i.e., the revised orders with abnormal data. The correction algorithm is as follows:

[0142] Based on the specific attribute values ​​of orders during peak and off-peak periods (such as order start time, end time, charging duration, transaction volume, charging cost, etc.), abnormal orders are corrected using prior rules and similarity algorithms. The correction algorithm is as follows:

[0143]

[0144]

[0145] In the formula, v′ i_abnormal These are the outlier values ​​in orders with data errors, i.e., the order values ​​that need to be corrected. These are normal values ​​in the data-abnormal bill, including charging volume, transaction amount, total meter reading, and total meter reading. To and The corresponding normal value in a normal bill; α is the proportional coefficient; v i_normal For with v′ i_abnormal The corresponding normal value in a normal bill.

[0146] Based on the order data in Step 1, we can derive the agreed-upon electricity prices and service fees for peak, off-peak, and valley periods for transaction order 247, as well as the specific charging time intervals. We can quickly calculate that the corresponding time interval for this bill is 2 hours and 16 minutes, and the corresponding service fee is 57.43 yuan. Therefore, the charging cost is 119.54 - 57.43 = 62.11 yuan. Thus, based on the price of peak electricity, the final amount of electricity charged for this order is 71.79 kWh.

[0147] 2) Missing abnormal orders

[0148] First, a predictive model is constructed to numerically correct for missing and abnormal orders. The specific steps are as follows:

[0149] The pre-trained model based on triples: Normal orders from vehicle network data are expressed as sentences in the form of "subject-verb-object", forming {e i_NO. / 1r i / 2 is / 3v i},in" / i " is the structure separator for the i-th sentence, "e i_NO. "r" i “v” i The sentence is divided into its main parts, and the sentence is input into the XLNet language model for sentence segmentation. The model then uses the search function to find the " / " symbols. i The position of "is used to obtain the triple e" i_NO. r i v i Each of the contents is grouped together, and the contents of the same type are used as a sequence as input for training the LSTM model.

[0150] The training model for a recurrent neural network is as follows: The input to the LSTM is a sequence S = {X}. <1> ,X <2> ,...,X <t>< / t> The sequence is of length T, and each element in the sequence is a sample point X = [x1, x2, ..., x]. n [], where n is the number of triples; in abnormal order data prediction, each sample data contains n features such as charging capacity and charging time, and T consecutive samples constitute a sequence. The abnormal order prediction value corresponding to the last element in the sequence is the label of the sequence, h is the hidden layer state, and each X corresponds to an output y, y ＜T＞ The calculation formula is:

[0151]

[0152]

[0153] W hy by Here, h represents the weights and biases of the output layer, g is the activation function (choose the appropriate activation function based on the specific situation), and h represents the weights and biases of the output layer. <t>< / t> This represents the hidden layer output function. Calculation using the forward input sequence Using the reversed input sequence for calculation, in order missing data prediction, y <t>< / t> This is the final predicted value corresponding to this sequence, where ||*||2 represents the L2 norm, and ||*|| 1 / 2 Indicate l 1 / 2 Norm, α represents the regularization coefficient, and the training fitted curve is as follows: Figure 5 As shown, the trained prediction model is obtained.

[0154] The trained prediction model is used to predict the missing relevant values ​​in abnormal orders, that is, to fill in the missing information in the electric vehicle settlement process, and to provide the corresponding charging time periods in peak, flat and valley periods, as well as user information and payment methods in the missing electricity bill.

[0155] The recurrent neural network prediction model uses the following method to predict missing or abnormal orders:

[0156] h′ t =f(W1x t +W3h′ t +b′ t (9)

[0157] h t =f(W2x) t +W4h t-1 +b t (10)

[0158]

[0159] y ＜T＞ =g(W hy ×h t +b y (12)

[0160] In the formula, x t This represents the input, specifically the numerical values ​​that are not missing from the core attributes; y <t>< / t> represents the output, i.e., the actual predicted missing values; f represents the activation function of the hidden layer of the recurrent neural network; W1, W2, W3, and W4 represent the first, second, third, and fourth power grid attribute association weight matrices after training through the recurrent neural network, respectively; h′ t Indicates the updated candidate state; h t-1 This represents the temporary value stored in the recurrent neural network for calculating the bill attributes at time t-1; h tThis represents the temporary value stored in the recurrent neural network for calculating the billing attributes at time t; b′ t This represents the bias vector updated by the recurrent neural network based on the bill attributes; b t H represents the bias vector fitted by the recurrent neural network based on bill attributes; t W represents the input state sequence; hy The output layer weight matrix is ​​represented by g; the activation function is represented by b. y This represents the bias matrix of the output layer; This is a vector concatenation operation.

[0161] The data for a certain charging station (No.: 1140290000011142) in a certain region in March is shown in the table below:

[0162]

[0163]

[0164] It can be found that bill number 55280 has missing order data, and some data for this order is 0;

[0165] After prioritizing the elimination of charging pile malfunctions and confirming that the charging pile is in an "on" state, the specific prediction steps are as follows:

[0166] S1. First, by calculating the total stop value of the electricity meter = 16629.35 - the total start value of the electricity meter = 16609.28, we can obtain the transaction electricity volume of this order as 20.07 kWh.

[0167] S2. Evaluate the specific charging time period based on the time difference between the order creation time and the next order, as well as the peak, flat, and valley values. Specifically, the charging period is from 14:59:53 on March 30, 2022 to 19:35:38 on March 30, 2022. That is, the vehicle should be charged during the flat period (15:00-18:00) or the peak period (18:00-21:00).

[0168] S3. Based on the charging curves of previous flat or peak charging periods, a recurrent neural network formula is applied, and the charging value is substituted to predict the most likely time period for that charging level. The specific formula is as follows:

[0169] h′ t =f(W1x t +W3h′ t +b′ t (13)

[0170] h t =f(W2x) t +W4h t-1 +bt (14)

[0171]

[0172] y ＜T＞ =g(W hy ×h t +b y (16)

[0173] Based on the charging time period types and electricity price information in the table below, we can make inferences.

[0174]

[0175] The process can further narrow down the range of the forecast by using the order creation time (2022-03-30 14:59:53). That is, the vehicle is most likely to be charged between 2022-03-30 15:00 and 2022-03-30 15:30:42. The corresponding peak, off-peak, and valley electricity consumption is 0, 0, 20.07, 0, and the transaction amount is 20.07 * 0.6538 = 13.13 yuan.

[0176] The data for a certain charging station (No.: 1140290000011142) in a certain region in March is shown in the table below:

[0177]

[0178]

[0179] It can be found that there is a difference in the meter reading between bill numbers 55219 and 55222. The specific difference in the meter reading is 70.07 kWh. After first checking that there is no charging pile fault, based on the start and stop status of the charging pile, it can be further confirmed that there are 2 missing bills.

[0180] The specific prediction steps are as follows:

[0181] S1. Based on the total stop value of the electricity meter = 16679.35 - the total start value of the electricity meter = 16609.28, the transaction electricity volume for the missing order difference is 70.07 kWh, including 10 kWh during peak hours and 60.07 kWh during off-peak hours;

[0182] S2. Applying the formula of a recurrent neural network, the total electricity value, the number of times the charging station has been started and stopped, the charging patterns of previous users of that station, and the account balance differences of related users are substituted into the model to predict the potential user ID and possible probability corresponding to the transaction electricity within the number of start and stop cycles. The specific formula is as follows:

[0183] h′ t =f(W1xt +W3h′ t +b′ t (17)

[0184] h t =f(W2x t +W4h t-1 +b t (18)

[0185]

[0186] y ＜T＞ =g(W hy ×h t +b y (20)

[0187] Ultimately, the prediction results are as follows:

[0188] 60.07kwh (User ID: xxx, probability: 0.60, User ID: yyy, probability: 0.25, User ID: zzz, probability: 0.14, ...)

[0189] 10.0 kWh (User ID: xxx, probability: 0.45, User ID: yyy, probability: 0.35, User ID: zzz, probability: 0.19, ...).

[0190] Based on the predicted probabilities, the data in the order with the highest probability is supplemented to obtain the revised missing and abnormal order.

[0191] Step 4: Data Upload

[0192] Normal orders and revised abnormal orders are uploaded to the vehicle networking platform or provincial application platform. The vehicle networking platform or provincial application platform then sends the corrected or normal invoice to the customer, who then proceeds with the order payment process.

[0193] This method can detect abnormal orders for electric vehicle charging stations and correct or predict the value of abnormal orders, thereby reducing economic losses caused by abnormal orders.

[0194] Example 2

[0195] This embodiment 2 provides a clearing and settlement device for abnormal electric vehicle charging pile orders, used to implement the clearing and settlement method for abnormal electric vehicle charging pile orders in embodiment 1, including:

[0196] Data acquisition module: used to acquire real-time orders from the vehicle networking platform;

[0197] Order Detection Module: Used to detect whether real-time orders are abnormal based on a preset inference engine, and to identify normal and abnormal orders;

[0198] Abnormal Order Revision Module: Used to perform numerical correction or data prediction on abnormal orders based on correction algorithms or preset prediction models to obtain revised abnormal orders;

[0199] Data upload module: Used to upload normal orders and revised abnormal orders to the vehicle networking platform.

[0200] Furthermore, the abnormal order revision module includes a data abnormal order revision module and a missing abnormal order revision module.

[0201] The data anomaly order revision module is used to perform numerical correction on data anomaly orders based on a correction algorithm to obtain revised data anomaly orders;

[0202] The missing and abnormal order revision module is used to perform numerical correction on missing and abnormal orders based on a preset prediction model, and obtain revised missing and abnormal orders.

[0203] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0204] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0205] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0206] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0207] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for clearing and settling abnormal orders for electric vehicle charging stations, characterized in that, include: Obtain real-time orders from the vehicle networking platform; The system uses a pre-set inference engine to detect whether the real-time order is an abnormal order. If the real-time order is a normal order, then the real-time order is directly uploaded to the vehicle networking platform; In response to the real-time order being an abnormal order, the real-time order is numerically corrected based on a correction algorithm, or the real-time order is predicted using a preset prediction model to obtain a revised abnormal order and upload it to the vehicle network platform. The method for constructing the inference engine includes: Obtain historical abnormal orders from the vehicle networking platform; Based on the aforementioned historical abnormal orders, and using an improved decision tree mining algorithm, the sorting and screening rules are obtained. Based on the mining sorting and filtering rules, combined with the custom sorting and filtering rules, an inference engine is obtained; The abnormal orders include orders with data errors and orders with missing data. In response to the fact that the abnormal order is a data abnormal order, the data abnormal order is numerically corrected based on the correction algorithm to obtain the revised data abnormal order and uploaded to the vehicle network platform; In response to the fact that the abnormal order is a missing abnormal order, the data of the missing abnormal order is predicted based on a preset prediction model to obtain a revised missing abnormal order and upload it to the vehicle network platform; The method for constructing the prediction model includes: Obtain historical normal orders from the vehicle networking platform and convert them into triplet form; Based on the pre-defined permutation language model XLNet, the historical normal orders are split into triples to obtain the input sequence; Based on the input sequence, a prediction model of a recurrent neural network is obtained by training a long short-term memory network. The expression for the prediction model is as follows: ； ； ； ； In the formula, x t This represents the input, specifically the numerical values ​​that are not missing from the core attributes; y <t>< / t> denoted by , representing the actual predicted missing values; f represents the activation function of the hidden layer of the recurrent neural network; W1, W2, W3, and W4 represent the first, second, third, and fourth power grid attribute association weight matrices after training through the recurrent neural network, respectively. Indicates the updated candidate state; This represents the temporary value stored in the recurrent neural network for calculating the bill attributes at time t-1; This represents the temporary value stored in the bill attribute calculation by the recurrent neural network at time t. This represents the recurrent neural network fitting and updating the bias vector based on bill attributes; This represents the bias vector fitted by the recurrent neural network based on bill attributes. Represents the input state sequence; This represents the output layer weight matrix; Indicates the activation function; This represents the bias matrix of the output layer; ⊕ is the vector concatenation operation; The expression for the improved decision tree mining algorithm is as follows: ； ； In the formula, Select the number of sorting and filtering rules for the inference engine; R i To comply with The number of abnormal orders in the sorting and filtering rules; D is the total number of abnormal orders; p i The support level of orders that meet the i-th sorting and filtering rule in the vehicle network data platform; This represents the Gini coefficient of each rule, and the rule with the smallest Gini coefficient among the selected rules is taken as the root node of the decision tree.

2. The method for clearing and settling abnormal electric vehicle charging pile orders according to claim 1, characterized in that, The expression for the correction algorithm is as follows: ； ； In the formula, These are the outlier values ​​in orders with data errors, i.e., the order values ​​that need to be corrected. These are normal values ​​in bills with abnormal data; To and The corresponding normal value in a normal bill; This is the proportionality coefficient; To and The corresponding normal value in a normal bill.

3. The method for clearing and settling abnormal electric vehicle charging station orders according to claim 1, characterized in that, The custom sorting and filtering rules include: The order start time is later than the end time. The transaction order volume is greater than 400; The initial value of the electricity meter in the transaction order is greater than the meter reading. The sum of the time-of-use electricity charges for each transaction order does not equal the total electricity charge. The sum of the electricity consumption of each transaction at different times does not equal the total electricity consumption. The sum of the time-based service fees for each transaction order does not equal the total service fee; The starting value of this transaction order is greater than the ending value of the previous order by more than 0.5; The number of times the charging completion status changed is inconsistent with the number of received transaction orders; There is no corresponding transaction order when the charging completion status change time value is compared with the charging order completion time value.

4. The method for clearing and settling abnormal electric vehicle charging station orders according to claim 1, characterized in that, Obtaining real-time orders from the connected vehicle platform includes: Obtain real-time order data from the vehicle networking platform; After unifying the semantic attributes of the real-time order data based on the knowledge fusion algorithm, the data is transformed into a triplet form to obtain the real-time orders of the vehicle network platform. The expression in the form of the triple is as follows: ； in, This represents the i-th order header entity. This represents the physical quantity of the order corresponding to the i-th category in the order. This represents the tail entity of the i-th order.

5. A clearing and settlement device for abnormal electric vehicle charging pile orders, used to implement the clearing and settlement method for abnormal electric vehicle charging pile orders as described in any one of claims 1-4, characterized in that, include: Data acquisition module: used to acquire real-time orders from the vehicle networking platform; Order detection module: used to detect whether the real-time order is an abnormal order based on a preset inference engine, and to obtain normal orders and abnormal orders; Abnormal Order Revision Module: Used to perform numerical correction or data prediction on the abnormal orders based on correction algorithms or preset prediction models to obtain revised abnormal orders; Data upload module: used to upload the normal orders and the revised abnormal orders to the vehicle networking platform.

6. The method for clearing and settling abnormal electric vehicle charging station orders according to claim 5, characterized in that, The abnormal order revision module includes a data abnormal order revision module and a missing abnormal order revision module; The data anomaly order revision module is used to perform numerical correction on data anomaly orders based on a correction algorithm to obtain revised data anomaly orders; The missing and abnormal order revision module is used to perform numerical correction on missing and abnormal orders based on a preset prediction model to obtain revised missing and abnormal orders.

Images