An adaptive matching power terminal exception work order dispatching method and system

Abstract

The disclosure provides a kind of self-adapting matching power terminal exception work order distribution method and system, comprising: obtaining power terminal related data, field processing personnel characteristic information and the historical record information of processing work order;According to the power terminal related data obtained, and the terminal exception fault identification model of pre-set, obtain exception work order class recognition result;According to the characteristic information of field processing class exception work order, field processing personnel characteristic information and the historical record information of processing work order, and the personnel-work order weight matching model of pre-set, obtain personnel-work order matching matrix, based on personnel-work order matching matrix, using KM algorithm realizes the distribution of personnel and to be distributed work order, output personnel-work order distribution result;The disclosure is based on considering field processing personnel characteristic, work order characteristic, historical processing record, using a kind of self-adapting matching power terminal exception work order distribution method, realizes the self-adapting matching and distribution of terminal exception work order.

Description

Technical Field

[0001] This disclosure belongs to the field of intelligent power information processing technology, and in particular relates to an adaptive matching method and system for dispatching abnormal work orders for power terminals. Background Technology

[0002] With the deepening of energy system and mechanism reform, the backbone of the future energy system will be electricity, and the proportion of electricity consumption will be further increased. As the execution equipment for electricity data collection, data management, bidirectional data transmission, and forwarding and execution of control commands, the power terminal is the foundation for ensuring a safe and stable supply of electricity.

[0003] The inventors of this disclosure have discovered that with the increase in the types and functions of power terminals, the types of work orders processed on-site have increased, and the volume of on-site work orders has increased. Traditional work order assignment methods have problems such as low efficiency, imbalance in personnel allocation, and low satisfaction with work order processing. At best, unsatisfactory work order processing results may lead to customer complaints; at worst, untimely processing of urgent work orders may lead to economic losses and personnel safety issues. All of the above problems are likely to generate public opinion risks and affect the process of power marketization and social electrification. Summary of the Invention

[0004] To address the aforementioned issues, this disclosure proposes an adaptive matching method and system for assigning abnormal work orders to power terminals. Based on considerations of on-site personnel characteristics, work order characteristics, and historical processing records, this disclosure employs an adaptive matching method for assigning abnormal work orders to power terminals, thereby achieving adaptive matching and assignment of abnormal work orders.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0006] Firstly, this disclosure provides an adaptive matching method for dispatching abnormal work orders for power terminals, including:

[0007] Acquire relevant data from power terminals, on-site personnel characteristics, and historical records of processing work orders;

[0008] Based on the acquired power terminal-related data and the preset terminal anomaly fault identification model, the abnormal work order type identification result is obtained; wherein, the terminal anomaly fault identification model is constructed using deep learning methods.

[0009] Remove abnormal work orders that can be repaired remotely and obtain abnormal work orders that can be handled on-site.

[0010] According to the feature information of the on-site processing type exception work order, the feature information of the on-site processing personnel, and the historical record information of the processed work order, and a preset personnel-work order weight matching model, a personnel-work order matching matrix is obtained, wherein the personnel-work order weight matching model is constructed by using a probability matrix decomposition algorithm;

[0011] Based on the personnel-work order matching matrix, a KM algorithm is used to realize the distribution of the personnel and the to-be-distributed work order, and a personnel-work order distribution result is output.

[0012] Further, the construction of the terminal abnormal fault identification model comprises:

[0013] A plurality of abnormal work order types are encoded as model output labels, and the preprocessed training set data is labeled according to the label type to obtain an input sample data set;

[0014] An LSTM algorithm is used to determine the type of the input sample set data work order;

[0015] A Softmax function is used as the last layer of the LSTM neural unit network training, and after the training of the gating device, a sample fault category prediction probability value is output, and a multi-class cross-entropy loss function is used to calculate the error between the expected output prediction probability value and the actual output probability of the sample data;

[0016] According to the obtained error, the parameters are iteratively updated by the gradient descent method, and when the loss function is lower than the error threshold, the optimal parameters are obtained;

[0017] After the iterative parameter adjustment and retraining process, the test set is used to test the prediction model, if the test result does not meet the model discrimination requirement, the error threshold is adjusted, and the parameter feedback is continued; if the test result meets the model discrimination requirement, the model parameters are fixed, and the terminal abnormal fault identification model is output.

[0018] Further, in the training of the terminal abnormal fault identification model, the normal sample set and the abnormal sample set in the initial sample set are divided into a training set, a test set and a validation set according to a preset proportion.

[0019] Further, the construction of the personnel-work order weight matching model comprises:

[0020] The length of service, skill and professional and area information of the on-site fault processing personnel are obtained, and the personnel attribute feature is defined;

[0021] The on-site processing work order history of the personnel is obtained, the number of different types of work orders processed by each personnel is determined, a personnel-work order processing record matrix is constructed based on the number of historical records, and the distance definition and cosine similarity algorithm are used to calculate the personnel processing record similarity;

[0022] Based on the field handling personnel dataset and the work order category feature dataset, a probability matrix decomposition algorithm is used to construct a personnel-work order weight matching model, and the processing weight value of the field handling personnel for different work order types is output.

[0023] When a new field handling personnel joins, based on the similarity of personnel feature attributes, the nearest neighbor personnel with a similarity less than a similarity threshold is selected, and the mean value of the personnel-work order weight value is obtained as the personnel-work order matching weight value of the new field handling personnel.

[0024] Further, the KM algorithm is used to realize the dispatching of personnel and to-be-assigned work orders, including:

[0025] Based on the personnel-work order matching matrix, the matching feature value of the personnel and the work order category is obtained, at this time, the existing idle personnel and to-be-assigned work orders are taken as vertices and are assigned values, the vertex of the personnel is assigned a maximum weight, and the vertex of the to-be-assigned work order is assigned a value of 0;

[0026] When matching, only edges with the same weight are matched, if no edge matching is found, the weight value of all personnel vertices of this path is reduced by a preset value, and the weight value of the to-be-assigned work order vertex is increased by the preset value, and then matching is performed again;

[0027] When the matching conflicts, only the personnel who have completed the matching are allowed to re-match with the work order with the least reduced expected value, so that the current personnel complete the matching;

[0028] The matching process is repeated until all personnel complete the work order matching.

[0029] Further, the region is divided according to the transformer area, according to the personnel-work order dispatching result, it is judged whether each subarea has field handling personnel, if yes, the work order dispatching information is issued according to the personnel-work order dispatching result, if not, the personnel-work order dispatching result is adjusted, and the work order dispatching information is issued under the premise of ensuring that each subarea can realize emergency event processing.

[0030] Further, based on the processing result of the field feedback abnormal work order, the historical record information of the processed work order is updated in a timely manner.

[0031] In a second aspect, the present disclosure also provides an adaptive matching power terminal abnormal work order dispatching system, including a data acquisition module, an abnormal work order category identification module, an abnormal work order processing module, a personnel-work order matching matrix acquisition module and a work order dispatching module.

[0032] The data acquisition module is configured to acquire power terminal related data, field handling personnel feature information and historical record information of processed work orders.

[0033] The abnormal work order type identification module is configured to obtain an abnormal work order type identification result according to the acquired power terminal related data and a preset terminal abnormal fault identification model, wherein the terminal abnormal fault identification model is constructed by a deep learning method.

[0034] The abnormal work order processing module is configured to eliminate the remotely repairable abnormal work order to obtain the on-site processing abnormal work order.

[0035] The personnel-work order matching matrix acquisition module is configured to obtain a personnel-work order matching matrix according to feature information of the on-site processing abnormal work order, feature information of the on-site processing personnel, and historical record information of processing work orders, and a preset personnel-work order weight matching model, wherein the personnel-work order weight matching model is constructed by a probability matrix decomposition algorithm.

[0036] The work order dispatching module is configured to realize dispatching of personnel and the to-be-dispatched work order based on the personnel-work order matching matrix by using a KM algorithm, and output a personnel-work order dispatching result.

[0037] In a third aspect, the present disclosure further provides a computer readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the self-adaptive matching power terminal abnormal work order dispatching method of the first aspect.

[0038] In a fourth aspect, the present disclosure further provides an electronic device including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the self-adaptive matching power terminal abnormal work order dispatching method of the first aspect when executing the program.

[0039] Compared with the prior art, the present disclosure has the following beneficial effects:

[0040] 1. The present disclosure aims to solve the problem that the current terminal abnormal work order processing needs to be checked on site to identify the work order category, collects terminal collection feature data based on historical work order processing records, and uses a time sequence recognition algorithm, i.e., an LSTM algorithm, to identify the specific work order category based on the changed collection feature data when the terminal normal operation becomes abnormal operation, thereby providing work order category information for dispatching of on-site personnel;

[0041] 2. The present disclosure is based on the matching problem of personnel and work orders, constructs a personnel-work order weight matching model based on personnel features and work order features by using a probability matrix decomposition algorithm, thereby giving a matching value of personnel and work order processing, ensuring the rationality of work order allocation based on seniority, distance, skills, and historical processing records, and giving a matching value of near neighbor personnel and work order processing based on personnel feature similarity when new personnel join, thereby solving the cold start problem;

[0042] 3. This disclosure adjusts the weight values ​​based on user satisfaction and achieves reasonable personnel scheduling in a relatively balanced manner while ensuring that each distribution area can handle emergency faults on site in a timely manner. Attached Figure Description

[0043] The accompanying drawings, which form part of this embodiment, are used to provide a further understanding of this embodiment. The illustrative embodiments and their descriptions are used to explain this embodiment and do not constitute an improper limitation of this embodiment.

[0044] Figure 1 This is an overall flowchart of Embodiment 1 of this disclosure;

[0045] Figure 2 This is a flowchart of the power terminal abnormal fault identification model constructed using the LSTM algorithm in Embodiment 1 of this disclosure;

[0046] Figure 3 The flowchart below shows the process of constructing a personnel-work order weight matching model using a probabilistic matrix factorization algorithm in Embodiment 1 of this disclosure.

[0047] Figure 4 This is a schematic diagram of the PMF model of Embodiment 1 of this disclosure;

[0048] Figure 5 This is a flowchart of the personnel and work order assignment model implemented using the KM algorithm in Embodiment 1 of this disclosure;

[0049] Figure 6 This is a diagram illustrating the personnel-work order assignment process of Embodiment 1 of this disclosure. Detailed Implementation

[0050] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0051] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0052] Example 1:

[0053] like Figure 1 As shown, this embodiment provides an adaptive matching method for dispatching abnormal work orders for power terminals, including:

[0054] The collected data from various terminals are obtained from the power system's production database. After sorting and summarizing the data, it is preprocessed to obtain an initial sample set.

[0055] Based on the historical terminal exception processing records, the corresponding historical collection data in the initial sample set is extracted to form an exception sample set and is labeled. According to the expert experience, the definition of various work order types and the processing priority weight are given. The normal sample set and the exception sample set in the initial sample set are divided into a training set, a test set and a verification set according to a ratio of n:m:1;

[0056] The training set sample feature data is taken as the input data in the training stage, and the LSTM algorithm is used to construct a terminal exception fault identification model to output an abnormal work order type identification result.

[0057] The remote repair work order is archived and prompt information is sent, and the remote repairable work order is excluded from the terminal exception work order type identification result, and the remaining on-site processing terminal exception work order type is obtained.

[0058] For the on-site processing terminal exception work order type, the on-site processing personnel feature information, the work order feature information and the historical records of processing work orders are taken as the input data, the PMF (Probabilistic Matrix Factorization) algorithm is used to construct a personnel-work order weight matching model, and the processing weight value of the on-site processing personnel to different work order types is output, that is, the personnel-work order matching matrix.

[0059] Based on the personnel-work order matching matrix, the KM algorithm is used to realize the dispatching of personnel and to-be-assigned work orders, and the personnel-work order dispatching result is output.

[0060] The region is divided according to the transformer area, the personnel-work order dispatching result is judged whether each partition has an on-site processing personnel, if yes, the work order dispatching information is issued according to the personnel-work order dispatching result, if not, the personnel-work order dispatching result is adjusted by manual intervention, and the work order dispatching information is issued under the premise of ensuring that each partition can realize emergency event processing.

[0061] Based on the processing result of the on-site feedback exception work order, the abnormal work order sample data of the initial sample set is updated in time.

[0062] As shown in Figure 2 In the embodiment, the terminal exception fault identification model is constructed by using the LSTM algorithm, which includes:

[0063] N kinds of abnormal work order types are encoded as model output labels, and the preprocessed training set data is labeled according to the label type to obtain input sample data set;

[0064] The LSTM algorithm is used to construct a terminal exception fault identification model, and the three control gates (input gate, forget gate and output gate) of the gate device in the LSTM algorithm are used to realize the discrimination of the input sample set data work order type.

[0065] The Softmax function is used as the last layer of the LSTM neural unit network training, and after the training of the gating device, the sample fault category prediction probability value is output, and then the multi-classification cross-entropy loss function is used to calculate the error (i.e. loss) between the expected output prediction probability value and the actual output probability of the sample data;

[0066] According to the calculated error back propagation, the parameters in the input sample set data are updated iteratively by gradient descent method, and when the loss function is lower than the error threshold, the relatively optimal parameters are obtained;

[0067] After the iterative parameter adjustment and retraining process, the test set is used to test the prediction model, if the test result does not meet the model discrimination requirement, go to the error threshold adjustment, continue to feedback the parameter adjustment; if the test result meets the model discrimination requirement, fix the model parameters, and output the terminal abnormal fault recognition model.

[0068] As shown in Figure 3 and Figure 4 In the present embodiment, a probability matrix decomposition algorithm is used to construct a personnel-ticket weight matching model, including:

[0069] The on-site fault handling personnel attribute characteristics are defined, the length of service, skill specialty and transformer area information of the on-site fault handling personnel are obtained from the marketing application system archive database, and the personnel attribute characteristics are defined;

[0070] The on-site handling ticket history records of the personnel are obtained from the marketing application system archive database, the handling quantity of different categories of tickets of each person is determined according to the history records, and the personnel-ticket handling record matrix is formed based on the history record quantity, and the distance definition and cosine similarity algorithm are used to calculate the personnel handling record similarity;

[0071] Based on the on-site handling personnel data set and the ticket category characteristic data set, a probability matrix decomposition algorithm is used to construct a personnel-ticket weight matching model, and the handling weight value of the on-site handling personnel for different ticket types is output;

[0072] When a new on-site handling personnel joins, based on the personnel characteristic attribute similarity, the nearest m persons less than the similarity threshold are selected, the mean value of the personnel-ticket weight value of the nearest m persons is obtained, and the mean value is used as the personnel-ticket matching weight value of the new on-site handling personnel.

[0073] As shown in Figure 5 Based on the personnel-ticket matching matrix, a KM algorithm is used to realize the allocation of the personnel and the to-be-allocated tickets, including:

[0074] Based on the personnel-ticket matching matrix, the matching characteristic value of personnel and ticket category is obtained, at this time, the existing idle personnel and the to-be-assigned ticket are taken as the vertex and the vertex value is assigned, the vertex value of the personnel is assigned as the maximum weight, and the vertex value of the to-be-assigned ticket is assigned as 0;

[0075] When matching, the matching principle is to match only with the edges of the same weight, if the edge matching cannot be found, the weight value of all personnel vertices of this path is reduced by a preset value d, the weight value of the to-be-assigned ticket vertex is increased by d, and then the matching is performed again;

[0076] When the matching conflicts, only the personnel who have completed the matching can be matched with the ticket with the least reduced expectation value, so that the current personnel complete the matching;

[0077] The matching operation is repeated until all personnel complete the ticket matching.

[0078] Embodiment 2:

[0079] As shown in Figure 1 , in this embodiment, a kind of adaptive matching power terminal abnormal order distribution method in the present disclosure is introduced according to specific examples, including:

[0080] A. Obtain the collected various terminal collection data from the production database of the power utilization information collection system, and after sorting and summarizing the data, pre-process the data to obtain an initial sample set;

[0081] Specifically, based on the field order processing records of a city from January 2020 to December 2020, about 100,000 field order processing samples are obtained; and based on the production database of the power utilization information collection system, the historical collection characteristic data of the terminals involved in the dispatched field processing order are obtained, thereby generating a field order processing sample set;

[0082] B. Based on more than 100,000 historical terminal abnormal processing records, the historical collection characteristic data of the terminals involved in the dispatched field processing order are obtained from the production database of the power utilization information collection system, including normal collection data and abnormal collection data, an initial sample set containing 100,269 samples, i.e. a field order processing sample set, is generated, and is divided into a training set, a test set and a validation set according to a 7:2:1 ratio; according to expert experience, the definition of various order types and the processing priority weight are given; part of the sample collection characteristic is shown in Table 1:

[0083] Table 1: Part of the terminal collection data characteristic example

[0084]

[0085]

[0086] C. Take 100269 data in the sample set as the training set sample feature data, take it as the input data of the training stage, and adopt the LSTM algorithm to construct the terminal abnormal fault identification model, and output the abnormal work order type identification result;

[0087] The step C, adopting the LSTM algorithm to construct the terminal abnormal fault identification model, has the following specific steps:

[0088] C1. Encode 10 types of abnormal work orders as model output labels, and label the preprocessed training set data according to the label type as input sample data set;

[0089] Among them, the 10 types of abnormal work orders include acquisition abnormal work order, metering abnormal work order, supplementary copy work order, on-site power-off and power-on, on-site power meter time correction work order, electric energy meter key download, electricity price adjustment, acquisition terminal key download, fee control work order and patrol work order;

[0090] C2. Adopt the LSTM algorithm to construct the terminal abnormal fault identification model, and according to the three control gates (input gate, forget gate and output gate) of the gate device in the LSTM algorithm, realize the identification of the input sample set data and the discrimination of the work order type:

[0091] The data processing process of the three control gates, memory information and output discrimination of the LSTM algorithm is:

[0092] Update the forget gate output:

[0093] f (t) =σ(W f h (t-1) +U f x (t) +b f )

[0094] Update the output of the two parts of the input gate:

[0095] i (t) =σ(W i h (t-1) +U i x (t) +b i )

[0096] a (t) =tanh(W a h (t-1) +U a x (t) +b a )

[0097] Memory information state:

[0098] C (t) =C (t-1) ⊙f(t) +i (t) ⊙a (t)

[0099] Update the output gate output:

[0100] o (t) = σ(W o h (t-1) + U o x (t) + b o )

[0101] h (t) = o (t) ⊙ tanh(C (t) )

[0102] Update the work order type discrimination prediction output of the current sample:

[0103]

[0104] Wherein, W f , W i , W a , W o and U f , U i , U a , U o are the weight parameters of the input hidden layer and the input sample respectively, t-1 is the processing data of the last LSTM neural unit, t is the processing parameter of the current LSTM neural unit, b f , b i , b a , b o are offset parameters;

[0105] C3. The Softmax function is used as the last layer of the LSTM neural unit network training. After the training of the gating device, the sample fault category prediction probability value is output, and then the multi-class cross-entropy loss function is used to calculate the error (i.e. loss) between the expected output prediction probability value and the actual output probability of the sample data.

[0106] The Softmax function is defined as follows:

[0107]

[0108] Wherein, y i is the prediction output value of the sample work order type discrimination, and C is the number of categories.

[0109] The cross-entropy loss function is defined as follows:

[0110]

[0111] wherein P is the expected output of the sample, T is the actual output of the sample, and C is the number of categories of classification, wherein T = [softmax(y1), softmax(y2), …, softmax(y i＝n )].

[0112] C4. The parameters of step C2 are iteratively updated by gradient descent method based on the error backpropagation calculated in step C3, and the relatively optimal parameters are obtained when the loss function is lower than the error threshold 0.02.

[0113] The model parameters are iteratively updated by gradient descent method, i.e. the partial derivatives of all parameters based on the loss function are calculated:

[0114]

[0115] wherein yi is the true label of the work order type i, pi is the predicted label of the work order type i calculated by the model through the Softmax function, k is the number of work order types, and N is the total number of samples. i i

[0116] C5. After the iterative parameter adjustment and retraining process, the relatively optimal model is obtained. The 21072 data from January to March 2021 obtained from the database are used as the test set to test the prediction model. If the test result does not meet the model discrimination requirement, go to step C4 to adjust the error threshold and continue feedback parameter adjustment. If the test result meets the model discrimination requirement, fix the model parameters and output the terminal abnormal fault identification model.

[0117] The matching accuracy of the terminal abnormal fault identification model is verified based on the 10640 verification sample set from April to May 2021. The matching results are shown in Table 2:

[0118] Table 2 Verification of matching accuracy of terminal abnormal fault identification model

[0119]

[0120]

[0121] Before iterative training:

[0122]

[0123] After iterative training:

[0124]

[0125] ​​D. Archiving and sending prompt information for remote repair work orders, and removing remote repairable work orders from the terminal exception work order type identification result, leaving the remaining on-site processing terminal exception work order type;

[0126] E. For on-site processing terminal exception work orders, use the on-site processing personnel feature information, work order feature information, and historical records of processing work orders as input data, use the Probabilistic Matrix Factorization (PMF) algorithm to construct a personnel-work order weight matching model, and output the processing weight values of on-site processing personnel for different work order types, i.e., the personnel-work order matching matrix;

[0127] The step E uses the Probabilistic Matrix Factorization algorithm to construct a personnel-work order weight matching model, and the specific steps are as follows:

[0128] E1. Definition of on-site fault processing personnel attribute characteristics

[0129] Obtain the length of service, skill specialty, and transformer area information of on-site fault processing personnel from the marketing application system archive database and define the personnel attribute characteristics;

[0130] Based on the length of service, the difference threshold of the length of service of different personnel is set to 1, and the length of service is rounded down when it does not meet 1. Let the length of service of personnel u be A u , and the definition of personnel length of service is as follows:

[0131]

[0132] The skill specialty category includes collection exception, metering exception, re-copied work order, on-site power restoration, on-site meter time correction work order, electric energy meter key download, electricity price adjustment, collection terminal key download, fee control work order, and patrol work order, and the definition is as follows:

[0133] T=[t0,t1,…,t N ]

[0134] Where t N takes 0 to represent not having the skill specialty, and takes 1 to represent having the skill specialty.

[0135] The definition based on the transformer area is as follows:

[0136] L=[l0,l1,…,l N ]

[0137] Where l N takes 0 to represent not being in one transformer area, and takes 1 to represent being in one transformer area.

[0138] E2. Obtain the processing order history records of each person from the marketing application system archive database respectively, determine the processing quantity of different types of orders of each person according to the records, and constitute a person-order processing record matrix based on the quantity, and calculate the processing record similarity of the persons by using the cosine similarity algorithm.

[0139] The cosine similarity algorithm calculation process is as follows:

[0140]

[0141] Wherein: S = {s1, s2, …, s g} represents the order type set, and the number of order types is g; N u,s and N v,s respectively represent the number of processing orders of the order type set s of the person u and the person v. Here, five persons are taken as an example, as shown in Tables 3a and 3b:

[0142] Table 3a Person-Order Processing Record Matrix

[0143]

[0144]

[0145] Table 3b Person-Order Processing Record Matrix

[0146]

[0147] Taking person 1 and person 2 as an example, the processing record similarity of the two persons can be calculated by the above cosine similarity algorithm to be about 0.7. The personnel attribute information is shown in Table 4:

[0148] Table 4 Personnel Attribute Information

[0149]

[0150]

[0151] The personnel with a small seniority gap are more likely to have similar processing experience. When the characteristic definition about seniority in step E1 is met, it is considered that the seniority attributes of the persons are the same, that is, the similarity is 1. The similarity calculation formula of the personnel based on seniority is as follows:

[0152]

[0153] The skill specialty also has an impact on the personnel. The skill specialty attribute can be used to measure the similarity of two persons. Let the skill specialty of the person u be T u , and the skill specialty of the person v be T v , then the skill specialty similarity of the two users is calculated as:

[0154]

[0155] Personnel location information is also an important factor to consider when dispatching abnormal work orders. Setting the threshold for the number of intervals between different work areas to 0, when the number of intervals between work areas is less than or equal to 0, it is considered that the location attributes of personnel in each work area are the same, i.e., the similarity is 1. We can set there to a total of n work areas (a region has 20 work areas), and let L be the number of personnel in the u-th work area. u There are personnel L in the vth district. v The formula for calculating the similarity of personnel based on their distribution across different transformer substations is as follows:

[0156]

[0157] Based on the above information, the final personnel feature similarity matrix can be obtained as follows:

[0158]

[0159] Where: α+β+δ=1; α,β,δ∈[0,1], representing the weights of different attributes respectively, here let α=0.2, β=0.4, δ=0.4.

[0160] Taking Person 1 and Person 2 as examples again, the similarity of their personnel feature information can be calculated by the above personnel feature similarity matrix, which is approximately 0.1.

[0161] Personnel similarity is calculated by combining the similarity of personnel processing records with the similarity of personnel feature information. The final calculation of personnel similarity is as follows:

[0162] Sem b (u,v)=γSem Label1 (u,v)+(1-γ)Sem Label2 (u,v)

[0163] Where γ represents the weighting parameter, which is set to 0.5 here. Then the similarity between person 1 and person 2 is...

[0164] Sem b (Personnel 1, Personnel 2) = 0.5 × 0.7 + 0.5 × 0.1 = 0.4.

[0165] E3. Calculation of Matching Feature Values ​​Between Personnel and Work Order Items

[0166] Assuming the abnormal work order dispatch system has M personnel feature information and N work order categories, then the matching matrix R between personnel and work order categories can be decomposed into the transpose U of the personnel latent feature matrix. T The product of the work order category latent feature matrix V and the elements R in the matching matrix R. i,jWorker i Type j U i and V j represent the column vectors of the feature matrices U and V, respectively. The PMF model is shown in Fig. 1. Figure 4

[0167] The PMF model assumes that each observation of the worker U i is independently and identically distributed, and the feature matrix U follows a Gaussian prior distribution with mean 0 and variance σ U 2 Then we have:

[0168]

[0169] Similarly, the feature matrix V also follows a Gaussian distribution with mean 0 and variance σ V 2 The Gaussian prior distribution of V satisfies the following equation:

[0170]

[0171] It is also assumed that the difference between the true value and the predicted value of the worker and the ticket type matching also follows a Gaussian distribution with mean 0 and variance σ R 2 Then the matching matrix R satisfies the following conditional probability:

[0172]

[0173] According to the Bayes formula, the posterior probability distribution of the latent feature matrices U and V is shown in the following equation:

[0174]

[0175] Maximizing the posterior probability is equivalent to minimizing the sum of the square error of the quadratic regularization term, and thus the objective function is obtained as follows:

[0176]

[0177] where λ U = σ R 2 / σ U 2 ; λ V = σ R 2 / σ V 2 ; and λ U and λ V are regularization coefficients.

[0178] ​E4. When new personnel join, select the nearest m people based on personnel similarity, and obtain the weight values of their personnel-ticket.

[0179] In the probability matrix decomposition model, the personnel similarity calculated in step E2 is fused to consider the influence of similar personnel on the potential feature vector of the personnel Worker i , and the potential feature vector of Worker i is represented as a weighted sum of the potential feature vectors of its similar personnel, and the calculation formula is:

[0180]

[0181] where M i represents the set of w similar neighbors of Worker i ; U e represents the potential feature vector of the similar neighbor e of Worker i ; P e,i represents the similarity between user Worker i and its similar neighbor e. After normalization processing P e,i , we have:

[0182]

[0183] Finally, output the matching matrix R, which is the matching between personnel Worker i and ticket type Type j .

[0184] F. Based on the personnel-ticket matching matrix, use the KM algorithm to realize the matching between personnel and to-be-assigned tickets, and output the personnel-ticket assignment result;

[0185] The step F uses the KM algorithm to realize the matching between personnel and to-be-assigned tickets, and the specific steps are as follows:

[0186] F1. In the above step E, the matching feature values of personnel and tickets have been calculated, at this time, the existing idle personnel and to-be-assigned tickets are taken as vertices and are assigned values, Figure 6 where the X set represents personnel and the Y set represents tickets. For any node in the graph, a function l is given to obtain the top index value of the node. In order to facilitate description, it is stipulated that l x (x) and l y (y) record the node top index values in set X and set Y respectively. Initially, the value of l x (x i ) is set as the maximum weight ω(x i ,y i ) of the edge e(x i ,y i ) associated with x i .i ), let l y (y i ) = 0, satisfying l x (x i )+l y (y i )≥ω(x i ,y i ), that is, the vertex of the personnel is assigned the maximum weight, and the vertex of the work order to be assigned is assigned 0;

[0187] F2. When matching, the matching principle is to match only the edges with the same weight. If no edge matching is found, modify the value of the feasible label of all personnel on this path. For the visited vertex x, subtract d from the feasible label of x:

[0188]

[0189] And for all visited vertices y, increase the feasible label of y by d, where S∈X, B∈Y;

[0190] F3. When the matching conflicts, only the personnel who have completed the matching can change a work order with the least decrease in the expected value to make the current personnel complete the matching;

[0191] F4. Repeat the operations of steps F2 and F3 until all personnel complete the work order matching.

[0192] G. Divide the area according to the transformer area, judge whether each subarea has a field handling personnel according to the personnel-work order assignment result. If yes, distribute the work order assignment information according to the personnel-work order assignment result. If no, manually intervene to adjust the personnel-work order assignment result, and distribute the work order assignment information under the premise of ensuring that each subarea can realize emergency event handling.

[0193] The field handling situation is shown in Table 5:

[0194] Table 5 Comparison of field handling situations

[0195]

[0196] Note: The work order processing satisfaction is 1-10 points, and the satisfaction is proportional to the score.

[0197] H. Based on the field feedback abnormal work order processing result, update the abnormal work order sample data of step B in time.

[0198] Example 3:

[0199] The embodiment provides a self-adaptive matching power terminal abnormal work order dispatching system, which comprises a data acquisition module, an abnormal work order type identification module, an abnormal work order processing module, a personnel-work order matching matrix acquisition module and a work order dispatching module.

[0200] The data acquisition module is configured to acquire power terminal related data, field processing personnel feature information and processing work order historical record information.

[0201] The abnormal work order type identification module is configured to obtain an abnormal work order type identification result according to the acquired power terminal related data and a preset terminal abnormal fault identification model, wherein the terminal abnormal fault identification model is constructed by a deep learning method.

[0202] The abnormal work order processing module is configured to eliminate remote repairable abnormal work orders and obtain field processing abnormal work orders.

[0203] The personnel-work order matching matrix acquisition module is configured to obtain a personnel-work order matching matrix according to the feature information of the field processing abnormal work orders, the feature information of the field processing personnel and the historical record information of the processing work orders and a preset personnel-work order weight matching model, wherein the personnel-work order weight matching model is constructed by a probability matrix decomposition algorithm.

[0204] The work order dispatching module is configured to realize the dispatching of personnel and to-be-dispatched work orders by using a KM algorithm based on the personnel-work order matching matrix, and output a personnel-work order dispatching result.

[0205] Embodiment 4

[0206] The embodiment provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the steps of the self-adaptive matching power terminal abnormal work order dispatching method in the embodiment 1.

[0207] Embodiment 5

[0208] The embodiment provides an electronic device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor realizes the steps of the self-adaptive matching power terminal abnormal work order dispatching method in the embodiment 1 when executing the program.

[0209] The above only describes the preferred embodiments of the present embodiment and is not used to limit the present embodiment. For those skilled in the art, the present embodiment can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present embodiment shall be included in the protection scope of the present embodiment.

Claims

1. A method for adaptive matching of power terminal exception work order dispatching, characterized in that, The method comprises the following steps: Obtain power terminal related data, field processing personnel characteristic information, and historical record information of processing work orders; Obtain an abnormal work order type identification result according to the obtained power terminal related data and a preset terminal abnormal fault identification model, wherein the terminal abnormal fault identification model is constructed by a deep learning method; Remove the abnormal work orders that can be remotely repaired to obtain field processing abnormal work orders; Obtain a personnel-work order matching matrix according to the characteristic information of the field processing abnormal work orders, the field processing personnel characteristic information, the historical record information of processing work orders, and a preset personnel-work order weight matching model, wherein the personnel-work order weight matching model is constructed by a probability matrix decomposition algorithm; Distribute the personnel and the to-be-assigned work orders by using a KM algorithm based on the personnel-work order matching matrix to output a personnel-work order distribution result, wherein the distribution of the personnel and the to-be-assigned work orders by using the KM algorithm comprises the following steps: Obtain matching characteristic values of the personnel and the work order types based on the personnel-work order matching matrix, at this time, existing idle personnel and to-be-assigned work orders are taken as vertices and are assigned values, the vertex of the personnel is assigned a maximum weight, and the vertex of the to-be-assigned work order is assigned 0; When matching, only edges with the same weight are matched, if no edge can be matched, the weight value of all personnel vertices of this path is reduced by a preset value, the weight value of the to-be-assigned work order vertex is increased by the preset value, and then matching is performed again; When a matching conflict occurs, only the personnel who have completed matching are allowed to re-match with the work order with the least reduced expected value, so that the current personnel complete matching; Repeat the matching process until all personnel complete work order matching.

2. A method for self-adapting matching power terminal exception work order dispatching as claimed in claim 1, wherein, The construction of the terminal abnormal fault identification model comprises the following steps: Encode multiple abnormal work order types as model output labels, and label the preprocessed training set data according to the label types to obtain input sample data sets; Determine the type of the input sample set data by using an LSTM algorithm; Take a Softmax function as the last layer of the LSTM neural unit network training, output sample fault category prediction probability values after the training of the gating device, and calculate the error between the expected output prediction probability values and the actual output probability of the sample data by using a multi-classification cross-entropy loss function; According to the obtained error, iteratively update the parameters by using a gradient descent method, obtain optimal parameters when the loss function is lower than an error threshold, and then perform re-training after the iterative parameter adjustment; After the iterative parameter adjustment and re-training, test the prediction model by using a test set, adjust the error threshold if the test result does not meet the model determination requirement, continue to feed back the parameter adjustment, fix the model parameters if the test result meets the model determination requirement, and output the terminal abnormal fault identification model.

3. A method of self-adapting matching power terminal exception work order dispatching as claimed in claim 2, wherein, During the training of the terminal abnormal fault identification model, normal sample sets and abnormal sample sets in an initial sample set are divided into a training set, a test set, and a verification set according to a preset ratio.

4. The method of claim 1, wherein the method further comprises: The construction of the personnel-work order weight matching model comprises the following steps: Obtain the length of service, skill specialty, and transformer area information of the field fault processing personnel, and define personnel attribute characteristics; Obtain the on-site processing work order history record of each personnel, determine the processing quantity of different types of work orders of each personnel, form a personnel-work order processing record matrix based on the historical record quantity, and calculate the personnel processing record similarity by using distance definition and cosine similarity algorithm; Based on the on-site processing personnel dataset and the work order category feature dataset, a personnel-work order weight matching model is constructed by using a probability matrix decomposition algorithm, and the processing weight value of the on-site processing personnel for different work order types is output. When a new on-site processing personnel joins, the nearest neighbor personnel with a similarity less than a similarity threshold is selected based on the personnel feature attribute similarity, and the mean value of the personnel-work order weight value of the nearest neighbor personnel is obtained as the personnel-work order matching weight value of the new on-site processing personnel.

5. The method of claim 1, wherein the method further comprises: The region is divided according to the transformer area, and based on the personnel-work order dispatching result, it is determined whether there is an on-site processing personnel in each subarea. If yes, the work order dispatching information is issued according to the personnel-work order dispatching result; if no, the personnel-work order dispatching result is adjusted, and the work order dispatching information is issued under the premise that each subarea can realize emergency event processing.

6. A method for self-adapting matching power terminal exception work order dispatching as claimed in claim 1, wherein, Based on the processing result of the on-site feedback abnormal work order, the historical record information of the processing work order is updated in a timely manner.

7. An adaptive matching power terminal exception work order dispatch system characterized by, The system comprises a data acquisition module, an abnormal work order type identification module, an abnormal work order processing module, a personnel-work order matching matrix acquisition module, and a work order dispatching module. The data acquisition module is configured to acquire power terminal related data, on-site processing personnel feature information, and historical record information of processing work orders. The abnormal work order type identification module is configured to obtain an abnormal work order type identification result according to the acquired power terminal related data and a preset terminal abnormal fault identification model, wherein the terminal abnormal fault identification model is constructed by a deep learning method. The abnormal work order processing module is configured to eliminate the remotely repairable abnormal work orders and obtain on-site processing abnormal work orders. The personnel-work order matching matrix acquisition module is configured to obtain a personnel-work order matching matrix according to the feature information of the on-site processing abnormal work orders, the on-site processing personnel feature information, the historical record information of the processing work orders, and a preset personnel-work order weight matching model, wherein the personnel-work order weight matching model is constructed by using a probability matrix decomposition algorithm. The work order dispatching module is configured to realize the dispatching of personnel and to-be-dispatched work orders by using a KM algorithm based on the personnel-work order matching matrix, and output the personnel-work order dispatching result. The realization of the dispatching of personnel and to-be-dispatched work orders by using the KM algorithm comprises: Based on the personnel-work order matching matrix, the matching characteristic values of personnel and work order categories are obtained, at this time, the existing idle personnel and to-be-dispatched work orders are taken as vertices and are assigned values, the vertex of the personnel is assigned a maximum weight, and the vertex of the to-be-dispatched work order is assigned 0; When matching, only the edges with the same weight are matched, if no edge matching is found, the weight value of all personnel vertices of this path is reduced by a preset value, and the weight value of the to-be-dispatched work order vertex is increased by the preset value, and then the matching is performed again. When the matching conflicts, only the completed matching personnel is matched with the work order with the least reduced expectation value, so that the current personnel completes the matching; The matching process is repeated until all personnel complete the work order matching.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program, when executed by a processor, implements the steps of the adaptive matching power terminal abnormal work order dispatching method of any one of claims 1-6.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor, when executing the program, implements the steps of the adaptive matching power terminal abnormal work order dispatching method of any one of claims 1-6.

Images