CN114444883A - Retail market risk monitoring, prevention and control early warning method for electricity selling company - Google Patents
Retail market risk monitoring, prevention and control early warning method for electricity selling company Download PDFInfo
- Publication number
- CN114444883A CN114444883A CN202111648252.1A CN202111648252A CN114444883A CN 114444883 A CN114444883 A CN 114444883A CN 202111648252 A CN202111648252 A CN 202111648252A CN 114444883 A CN114444883 A CN 114444883A
- Authority
- CN
- China
- Prior art keywords
- value
- risk
- electricity
- retail
- early warning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000005611 electricity Effects 0.000 title claims abstract description 159
- 238000000034 method Methods 0.000 title claims abstract description 75
- 238000012544 monitoring process Methods 0.000 title claims abstract description 28
- 230000002265 prevention Effects 0.000 title claims abstract description 11
- 238000004364 calculation method Methods 0.000 claims abstract description 63
- 238000010845 search algorithm Methods 0.000 claims abstract description 11
- 238000013507 mapping Methods 0.000 claims description 105
- 230000006870 function Effects 0.000 claims description 54
- 238000012545 processing Methods 0.000 claims description 47
- 238000010606 normalization Methods 0.000 claims description 19
- 238000010248 power generation Methods 0.000 claims description 13
- 230000002159 abnormal effect Effects 0.000 claims description 12
- 238000012888 cubic function Methods 0.000 claims description 12
- 238000012887 quadratic function Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 11
- 230000001276 controlling effect Effects 0.000 claims description 9
- 230000001960 triggered effect Effects 0.000 claims description 6
- 238000004140 cleaning Methods 0.000 claims description 5
- 108010074864 Factor XI Proteins 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 4
- 238000011084 recovery Methods 0.000 claims description 4
- 230000001105 regulatory effect Effects 0.000 claims description 4
- 239000000725 suspension Substances 0.000 claims description 4
- 230000003827 upregulation Effects 0.000 claims description 4
- 238000013112 stability test Methods 0.000 claims description 3
- 230000002354 daily effect Effects 0.000 claims 9
- 230000003203 everyday effect Effects 0.000 claims 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims 1
- 238000005457 optimization Methods 0.000 abstract description 21
- 238000011161 development Methods 0.000 abstract description 6
- 238000013136 deep learning model Methods 0.000 abstract description 3
- 238000012360 testing method Methods 0.000 description 12
- 238000009826 distribution Methods 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000012163 sequencing technique Methods 0.000 description 6
- 238000012954 risk control Methods 0.000 description 5
- 238000000354 decomposition reaction Methods 0.000 description 4
- 238000009792 diffusion process Methods 0.000 description 4
- 238000009827 uniform distribution Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 238000001278 Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000005295 random walk Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000003646 Spearman's rank correlation coefficient Methods 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 1
- 239000010931 gold Substances 0.000 description 1
- 229910052737 gold Inorganic materials 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- General Business, Economics & Management (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Entrepreneurship & Innovation (AREA)
- Computational Linguistics (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Public Health (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Educational Administration (AREA)
- Evolutionary Biology (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a retail market risk monitoring, prevention and early warning method for an electricity selling company, and belongs to the technical field of electricity selling settlement risk monitoring. According to the method, historical power consumption data of users of the power selling company are automatically acquired, the super-parameter optimization is carried out on a deep learning model by using a random fractal intelligent search algorithm, and then monthly transaction capacity of the power selling company is predicted based on the model, so that effective reference is provided for accurate control of retail transaction prediction capacity of the power selling company. Meanwhile, based on the predicted retail transaction capacity, a self-adaptive daily transaction risk grading early warning prevention and control mode is provided in consideration of a zero-price difference calculation mode of the current retail electric power market, the early warning prevention and control mode is linked with a power consumption value, corresponding early warning grades and early warning measures are automatically matched, early warning references are provided for risk management of power grid enterprises, the operation settlement risks of power grid companies can be effectively prevented and controlled, the market operation efficiency is optimized, and the stable and ordered development of multi-market main bodies in the retail electric power market is guaranteed.
Description
Technical Field
The invention belongs to the technical field of electric sales settlement risk monitoring, and particularly relates to a risk monitoring, prevention and control early warning method for retail markets of electric sales companies.
Background
In recent years, with the continuous deepening of the electric power market reformation, the vitality of the market main body at the electricity selling side is continuously enhanced, and important support is provided for the flourishing development of the electric power market. However, on the electricity selling side, emerging electricity selling companies also face multiple uncertain factors such as market user demand change, electric energy price fluctuation and unstable self-operation conditions, and further cause market risks. Meanwhile, as the market openness degree is continuously increased, the settlement main body of the retail electric power market is changed, under the background of the existing market environment and policies, the settlement risk of the power grid company in the retail market is changed from the single main body settlement risk of the power selling company-the power grid company into the multi-main-body settlement risk coupled with the multiple settlement risks of the power plant-the retail user-the power selling company-the power grid company, the settlement risk is greatly increased, higher requirements are put on the settlement risk control capability of the power grid company, and meanwhile, the control of the settlement risk and the development of the electric power market are further restricted by the relative shortage of the existing market risk monitoring technology. Therefore, how to overcome the defects of the prior art is a problem which needs to be solved urgently in the technical field of the current electricity selling settlement risk monitoring.
Disclosure of Invention
Aiming at the current situation that the settlement risk of a power grid enterprise in the current retail electric power market is changed from single-main-body settlement risk to multi-main-body settlement risk coupled with multiple risks, and the problems of information lag, insufficient main-body benefit guarantee dimension, lack of effective risk control measures and the like of the power grid enterprise in the traditional risk management, the invention designs the retail market risk monitoring, controlling and early warning method of the power selling company combined with the deep learning technology, which takes the income difference of wholesale price-retail price as a judgment condition, intelligently optimizes the multi-level risk control measures based on the rights and interests guarantee visual angle of a central counterparty of the power grid company, reduces the operation settlement risk of the power grid company, optimizes the market operation efficiency, improves the market operation guarantee capability, and promotes the stable and orderly development of the multi-market main body in the retail electric power market.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a retail market risk monitoring, prevention and control early warning method for an electricity selling company comprises the following steps:
the method comprises the following steps that (1) historical electricity consumption data of a user corresponding to an electricity selling company, and temperature, humidity and day type data of a place where the user is located are obtained;
step (2), arranging the historical electricity consumption data obtained in the step (1) according to time sequence to serve as a historical electricity consumption time sequence, carrying out normalization processing, and then decomposing the normalized historical electricity consumption time sequence into K sub-modal components by adopting a VMD method;
step (3), carrying out nonlinear mapping processing on the temperature, humidity and day type data obtained in the step (1);
step (4), taking the value of the normalized historical power consumption quantum modal component in the previous B months, the temperature, humidity and day type data subjected to the nonlinear mapping processing in the previous B months as the input of the LSTM model, taking the value of the normalized historical power consumption quantum modal component in the current month as the output of the LSTM model, and training the LSTM model of each sub-modal component to obtain the LSTM power consumption prediction model of each sub-modal component; b is more than or equal to 1;
step (5), historical electricity consumption is calculatedThe value of the sub-model component in the previous B months, the temperature, the humidity and the day type data processed by the nonlinear mapping of the B months are used as input, the input is input into the LSTM power consumption prediction model of the corresponding component obtained in the step (4), the normalized power consumption prediction value of each component is obtained, then the normalized power consumption prediction value of each component is subjected to reverse normalization, accumulation is carried out, and the current-month power consumption prediction value Q is obtainedcapacity；
Step (6), according to the predicted value Q of the electricity sold in the monthcapacityCalculating zero price difference delta P of retail market batches of daily electricity selling companies;
and (7) tracking the retail transaction risk of the market based on the daily calculated retail market batch zero-valent difference calculation result delta P, and matching the corresponding early warning levels and early warning measures to finish the retail market risk tracking and early warning.
Further, it is preferable that, in the step (1), the obtained historical power consumption data is subjected to a stationarity check, and if the assumed result is rejected, the data is subjected to data cleansing.
Further, preferably, the specific method for data cleaning is as follows: processing abnormal values and missing values; and taking the abnormal value as a missing value, uniformly processing the two types of abnormal values, and interpolating the missing value by adopting a quadratic spline interpolation method.
Further, it is preferable that, in the step (1), the types of days include working days, saturdays, sundays and holidays in four seasons of spring, summer, autumn and winter.
Further, preferably, in the step (3), a specific method of the nonlinear mapping processing of the temperature, humidity and day type data is as follows:
1) for the temperature:
the lower temperature threshold is set astThe upper limit temperature threshold is set to
Setting the temperature factor tnThe pre-mapping value is xnThe mapped value is ynWhen is coming into contact with
Then, a normalization method is adopted for processing; when in use
Or t<tAnd then, processing by adopting a mapping relation function:
firstly, the average value of the user electricity consumption corresponding to each temperature is counted in the obtained user historical electricity consumption data set, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the temperature-electricity consumption average value, the fitting goodness of three fitting curves and the before-mapping value is respectively calculated, namely, the coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein x isiFor the value before the i-th mapping,
represents the mean of the values before the mapping,
represents the ith mapped value;
after the fitting relation curve is determined, inputting a pre-mapping value to calculate an application electric quantity value, and normalizing the value in a corresponding interval to obtain a mapped value;
2) for the humidity:
the lower limit humidity threshold is set asrThe upper limit humidity threshold is set to
Setting the humidity factor rnThe pre-mapping value is unMapped value is wnWhen is coming into contact with
When the method is used, a normalization method is adopted for processing, and when r is used, the method is used for processingn＜r0When is coming into contact with
Or r<rAnd then, processing by adopting a mapping relation function:
firstly, the average value of the power consumption of the user corresponding to each humidity in the obtained historical power consumption data set of the user is counted, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the humidity-power consumption average value, the fitting goodness of three fitting curves and a mapping pre-value is respectively calculated, namely, a coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein u isiFor the value before the i-th mapping,
represents the mean of the values before the mapping,
represents the ith mapped value;
after the fitting relation curve is determined, inputting a pre-mapping value to calculate an application electric quantity value, and normalizing the value in a corresponding interval to obtain a mapped value;
3) for day type data:
day type data fnMappingThe front value is hnThe mapped value is znAnd processing by adopting a mapping relation function:
firstly, the average value of the user power consumption corresponding to each day type is counted in the obtained user historical power consumption data set, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the day type-power consumption average value, the fitting goodness of three fitting curves and a mapping pre-value is respectively calculated, namely, a coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein h isiFor the value before the i-th mapping,
represents the mean of the values before the mapping,
representing the ith mapped value.
Further, preferably, in the step (4), a random fractal search algorithm SFS is adopted to perform optimization solution on LSTM parameters, wherein the parameters include the number of hidden layer nodes, the time step and the learning rate;
in the step (2), K in the VMD method is determined by adopting a random fractal search algorithm SFS.
Further, preferably, in step (6), the specific method for calculating the zero-price difference of the daily retail market lot is as follows:
calculating and calculating the zero-price difference of the daily retail market batch according to the latest daily wholesale transaction and retail transaction data
Wherein,
indicating the retail income of the electricity-selling company,
indicating the wholesale cost of the electricity selling company,
the calculation method is as follows:
wherein N isretailIndicating the number of retail customers of the electricity-selling company,
represents the retail market transaction price of the electricity consumer i,
representing the electricity consumption of the retail market of the electricity consumer i;
wherein Q iscapacityIndicating the predicted value of the electricity usage in the current month,
electric selling deviceThe electricity consumption is wholesale in the month of the department,
indicating the current monthly wholesale transaction average price of the electricity selling company, PdIndicating the monthly deviation electricity reference price, xioShowing the penalty factor xi of the electricity selling company for the excess electricityuPenalty coefficient, Q, indicating low power consumption for electricity selling companyoverThe calculation value of the potential electricity overuse amount of the electricity selling company is represented by the following calculation method:
Quthe calculation value of the potential power consumption shortage of the power selling company is represented by the following calculation method:
Qunderthe method is characterized in that a potential scarce assessment electric quantity measurement value of an electricity selling company is represented, and the calculation mode is as follows:
deviation electric quantity reference price PdThe calculation formula is as follows:
in the formula, QClean and superhairFor clean energy power plant monthly total clean super power generation quantity, QThermal power super power generationFor the monthly integral net excess power generation quantity, P, of the thermal power plantThermal power hanging plateFor regulating the electric quantity for thermal power, the price of the tag is unified, P0And showing the up-regulation service reference price of the region.
Further, preferably, in the step (7), the retail market transaction risk is tracked for the daily retail market batch zero-valent difference calculation result, and the corresponding early warning level and early warning measure are matched, and the specific method is as follows:
if the daily retail market batch zero-price difference is less than 0 and the loss of retail income reaches the performance guarantee amount alphariskIn which α isriskIf the current is less than 1, triggering first-level early warning, and sending an operation attention letter to the power selling company by the power grid company to remind the power selling company to pay attention to self settlement risk;
if the daily retail market batch zero-price difference is less than 0, and the loss of retail income reaches the beta of the performance guarantee fund amountriskIn which α isrisk＜βriskIf the number is less than 1, triggering second-stage early warning, and sending an operation inquiry letter to the power selling company by the power grid company;
if the daily retail market batch zero-price difference is less than 0, and the loss of retail income reaches the gamma of the performance guarantee fund amountriskIn which phi isrisk≤γrisk＜1Uφrisk＞βriskTriggering a third-level early warning, sending a risk warning letter to the power selling company by the power grid company, taking a settlement suspension measure, and implementing the settlement recovery according to the next day after the completion of the subsidy payment and performance guarantee;
if the daily retail market batch zero-price difference is less than 0, and the loss of retail income reaches gamma of the performance bond guarantee fund amountriskWherein r isriskAnd (3) triggering a fourth-level early warning, taking a fusing measure by a power grid company, immediately taking a settlement suspending measure for an electric selling company, and deducting a market credit point P of the electric selling companyscoreWhen P isscore≥NriskThen, the corresponding market admission settlement period limiting measures are carried out on the electricity selling company, namely, the time period is separated after the completion guarantee fund is subsidized
The later process can settle accounts with the power grid company again; wherein N isriskRepresenting a market credit point risk threshold, alpha, set by the grid companyrisk、βrisk、γrisk、ΓriskRespectively represents the proportion of the retail yield loss to the performance bond guarantee fund when the first, second, third and fourth-level early warning are triggered, phiriskShow the contribution of retail profit loss when triggering the third-level early warningApproximately guarantee the lower limit ratio of gold.
The normalization process is not particularly limited in the present invention, and a maximum-minimum normalization method is preferably used.
In the invention, the mapping relation function is determined by taking the determined coefficient closest to 1 as an index, namely the determined coefficient of which function of the three functions is closest to 1, and the function is the mapping relation function.
The invention combines the current situation of market settlement, starts from the perspective of guaranteeing the rights and interests of a central counterparty of a power grid company, sets up the benefits of all parties of a market main body, designs a risk monitoring, preventing and early warning method for the retail market of the power selling company based on a deep learning technology, introduces a random fractal search intelligent optimization algorithm as a super-parameter optimization mode based on the electric quantity prediction of electric power retail users considering a plurality of influence factors, provides complete full-flow refined multi-level risk monitoring of the electric power retail trade market 'prediction-settlement-early warning', effectively reduces the settlement risk of the retail electric power market, and provides effective support for the risk control of the electric power market.
Compared with the prior art, the invention has the beneficial effects that:
according to the method, the super-parameter optimization is performed on the deep learning model by using the random fractal intelligent search algorithm based on the historical power consumption data of the users of the power selling company automatically acquired by the system, and then the monthly transaction capacity of the power selling company is predicted based on the model, so that effective reference is provided for accurate control of the retail transaction prediction capacity of the power selling company. Meanwhile, based on the predicted retail transaction capacity, a current retail electric power market batch zero-price difference calculation mode is considered, a self-adaptive fine daily transaction risk grading early warning prevention and control mode is provided, the mode is linked with a power consumption predicted value, corresponding early warning grades and early warning measures are automatically matched, early warning references are provided for power grid enterprise risk management, the operation settlement risk of a power grid company can be effectively prevented and controlled, the market operation efficiency is optimized, and the stable and ordered development of multi-market main bodies in the retail electric power market is guaranteed.
Drawings
FIG. 1 is a schematic diagram of a VMD-LSTM-SFS algorithm model in the present invention;
fig. 2 is a schematic flow chart of the risk monitoring, prevention, control and early warning method for the retail market of the electricity selling company.
Detailed Description
The present invention will be described in further detail with reference to examples.
It will be appreciated by those skilled in the art that the following examples are illustrative of the invention only and should not be taken as limiting the scope of the invention. The examples do not specify particular techniques or conditions, and are performed according to the techniques or conditions described in the literature in the art or according to the product specifications. The materials or equipment used are not indicated by manufacturers, and all are conventional products available by purchase.
Problem definition
With the deep advance of the reform of the electric power system, the reform of the settlement of the power selling companies and the power grid is orderly advanced all over the country. In the original mode, the electric power selling company and each market main body in the retail electric power market settle accounts in a mode that the electric power selling company settles the electric power transmission and distribution fee, the retail electric energy fee and the wholesale electric energy fee with the power grid company, the retail electric power users and the power plant respectively, the electric power selling company is in a core settlement position, the electric power grid company only has the retail settlement relationship of the electric power transmission and distribution price with the electric power selling company, and the electric power grid retail settlement risk is smaller and more single. In a new mode, the state clearly requires that a power grid company and market main bodies such as an electricity selling company, a retail user, a power plant and the like perform power price and electricity price settlement, the power grid enterprise performs settlement with the electricity selling company after uniform collection, the retail market settlement main body is changed from a single electricity selling company into a multi-market main body, settlement risks are greatly increased, the conventional settlement risk supervision measures mainly take monthly batch zeroth order risk monitoring as a main measure, the monthly batch zeroth order is calculated in a trial mode in the beginning of the next month to determine final batch zeroth order, and further the risk control is performed on the electricity selling company. The monitoring time scale is mainly monthly, daily risk monitoring is assisted, monthly risks can be better controlled by quantitative calculation, but the problem of overlong time period exists, the daily staring risk monitoring can shorten the monitoring time scale, the monitoring efficiency is greatly increased, and the market operation stability is improved. Therefore, on one hand, for the electricity selling company, the accurate prediction of the monthly transaction capacity is the key to whether the transaction mechanism can accurately calculate the batch zeroth order difference yield, and further the self power grid settlement risk is controlled; on the other hand, for power grid enterprises, the accuracy of zero price difference and risk prevention, control and early warning under a refined time scale are the key points for managing market risks. Specifically, in the invention, based on historical power consumption data of users of the power selling company automatically acquired by the system, a random fractal intelligent search algorithm is used for carrying out hyper-parameter optimization on a deep learning model, and then monthly transaction capacity of the power selling company is predicted based on the model. The core flow of the technical scheme is as follows:
acquiring historical power consumption data of a user corresponding to an electricity selling company, and temperature, humidity conditions and day type data of a place where the user is located;
the retail transaction system of the transaction mechanism obtains historical electricity consumption data of the electricity selling company in the beginning of each month, the system can obtain corresponding data of the electricity selling company independently or in batches, and the data of the electricity selling quantity of each user of the corresponding electricity selling company is obtained through automatic matching according to the number of the user of the agent of the electricity selling company. Meanwhile, the system automatically collects the temperature and humidity conditions and the day type data of the local city where the user is located, and stores the data into the database. And (4) performing stability test on the obtained historical power consumption data according to the obtained historical power consumption data, and selecting a KPSS test method to test the data stability. If the assumed result is rejected, i.e. the time sequence is not stable or the trend is not stable, the data is subjected to data cleaning, and abnormal values and missing values are processed firstly. And taking the abnormal value as a missing value, uniformly processing the two types of abnormal values, and interpolating the missing value by adopting a quadratic spline interpolation method.
Arranging the obtained historical electricity consumption data according to time sequence to serve as a historical electricity consumption time sequence and carrying out normalization processing, and then decomposing the normalized historical electricity consumption time sequence into K sub-modal components by adopting a VMD method;
processing the normalized time sequence by using a non-recursive and variational modal decomposition method by using a VMD (virtual machine tool), and decomposing the time sequence f into K submodes, wherein K is determined by a random fractal search algorithm (SFS) in step (four) as a hyper-parameter (preferably K is 15), namely a finite bandwidth modal function { u } uk(t) }, K1, 2, 3, K, for each sub-modal function uk(t), calculating corresponding analytic signals by using Hilbert transform, and obtaining corresponding single-side frequency spectrums as follows:
for each mode function uk (t), the center frequency ω of each single-component am fm signal corresponding theretokIndex term of (2)
Aliasing, changing to its corresponding base band:
solving the signal bandwidth of the submode according to a Gaussian smoothing method for mediating signals, wherein an objective function is as follows:
wherein, { uk}＝{u1，...，uK}；{ωk}＝{ω1，...，ωK}；{uk}、{ωkCorrespondingly decomposing the kth modal component and the center frequency respectively; δ (t) is a dirac function; the operation is convolution, t represents time, and j is an imaginary unit;
and converting the problem into an unconstrained variational optimization problem by adopting a secondary penalty factor alpha and a Lagrange multiplier lambda (t):
pair sub-mode
Frequency of
And lagrange multiplier λn+1And performing alternate updating, wherein the final solution is as follows:
wherein,
respectively represent
f(t)、ui(t), λ (t), i ≠ k, n denotes the number of iterations.
Setting the precision epsilon of the termination condition of the iterative computationbCalculation accuracy εcThe following were used:
if epsilonc＜εbThe model stops iterating, otherwise, the loop calculation is continued.
Thirdly, carrying out nonlinear mapping processing on the obtained temperature, humidity and day type data;
because the temperature, humidity and day type data have great influence on the electricity consumption of the user, for example, under general conditions, the electricity consumption in weekends and holidays is obviously smaller than that in working days and electricity consumption, the electricity consumption in hot seasons or cold seasons with high temperature has great influence on the electricity consumption of the user, and the unit dimensions of all the influence factors are different, and therefore, a nonlinear mapping method is adopted to map all the different influence factors to a certain interval.
Firstly, the temperature is divided into three categories of higher, moderate and lower, and the lower limit temperature threshold of the category boundary is set astThe upper limit temperature threshold is set to
Setting the temperature factor tkThe pre-mapping value is xkThe mapped value is ykThen the configurable non-linear mapping relationship vector is [ t ]k，xk，yk]. Similarly, the humidity is divided into three categories of higher humidity, moderate humidity and lower humidity, and the lower humidity threshold of the category is set asrThe upper limit humidity threshold is set to
Setting the humidity factor rkThe pre-mapping value is ukMapped value is wkThen the configurable non-linear mapping relationship vector is [ rk，uk，wk]. For the daily type, because the daily electricity consumption of the users is not uniformly distributed in the year and the seasonal influence is obvious, i interval dimensions are firstly divided in the year, the electricity consumption behaviors of the users are clustered and distributed in different time intervals, the characteristics are described in each interval, and the characteristic labels are given
Such as weekdays, weekends, and holidays. Therefore, the nonlinear mapping relation vector formed by the mapping pre-value dk and the mapping post-value zk of each feature in the interval is set as
The mapping relationship is determined as follows:
1) for the temperature:
the lower temperature threshold is set totThe upper limit temperature threshold is set to
Setting the temperature factor tnThe pre-mapping value is xnThe mapped value is ynWhen is coming into contact with
Then, a normalization method is adopted for processing; when in use
Or t <tWhen (t in Table 1)1～tnDEG C), processing by adopting a mapping relation function:
TABLE 1
Firstly, the average value of the user electricity consumption corresponding to each temperature is counted in the obtained user historical electricity consumption data set, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the temperature-electricity consumption average value, the fitting goodness of three fitting curves and the before-mapping value is respectively calculated, namely, the coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein xi is the ith pre-mapping value,
represents the mean of the values before the mapping,
represents the ith mapped value;
after the fitting relation curve is determined, inputting a pre-mapping value to calculate an application electric quantity value, and carrying out normalization processing on the value in a corresponding interval to obtain a post-mapping value;
2) for the humidity:
the lower limit humidity threshold is set asrThe upper limit humidity threshold is set to
Setting the humidity factor rnThe pre-mapping value is unMapped value is wnWhen is coming into contact with
When the method is used, a normalization method is adopted for processing, and when r is used, the method is used for processingn＜r0When is coming into contact with
Or r <rWhen (r in Table 2)1～rn) And processing by adopting a mapping relation function:
TABLE 2
Firstly, the average value of the power consumption of the user corresponding to each humidity in the obtained historical power consumption data set of the user is counted, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the humidity-power consumption average value, the fitting goodness of three fitting curves and a mapping pre-value is respectively calculated, namely, a coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein ui is the ith pre-map value,
represents the mean of the values before the mapping,
represents the ith mapped value;
after the fitting relation curve is determined, inputting a pre-mapping value to calculate an application electric quantity value, and normalizing the value in a corresponding interval to obtain a mapped value;
3) for day type data:
day type data fnThe pre-mapping value is hnThe mapped value is znAnd processing by adopting a mapping relation function:
TABLE 3
Wherein,
respectively showing the working day, saturday, sunday and holiday (calculated according to the longest day of 7) in four seasons of spring, summer, autumn and winter.
Firstly, the average value of the user electricity consumption corresponding to each day type (as shown in table 3) is counted in the obtained user historical electricity consumption data set, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the day type-electricity consumption average value, the goodness of fit of three fitting curves and the value before mapping is respectively calculated, namely, the coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein h isiFor the value before the i-th mapping,
represents the mean of the values before the mapping,
representing the ith mapped value.
Dividing a data set into a training set and a testing set, optimizing the number of hidden layer nodes, time step length and learning rate of the LSTM by using a random fractal search algorithm SFS, substituting an optimal parameter solution set into each sub-mode training set of the LSTM through an optimization algorithm, and establishing a VMD-SFS-LSTM sub-mode prediction model;
and dividing a data set (including historical electricity consumption data, temperature, humidity and day type data) D into a training set M and a testing set T according to a 6: 4 ratio, wherein M, T test distribution is consistent with D so as to improve the adaptability of a subsequent model. Training by using a training set M to construct an LSTM power consumption prediction model; the method specifically comprises the following steps:
taking the value of the normalized historical power consumption quantum modal component in the previous A month, the temperature, the humidity and the day type data subjected to nonlinear mapping processing in the A month as the input of an LSTM model, taking the value of the normalized historical power consumption quantum modal component in the current month as the output of the LSTM model, and training the LSTM model of each sub-modal component to obtain an LSTM power consumption prediction model of each sub-modal component; a is more than or equal to 1;
when an LSTM neural network model is established, a random fractal search algorithm SFS is used for carrying out optimization solution on LSTM parameters, wherein the parameters comprise hidden layer node numbers, time step lengths, learning rates and sub-module component numbers K, the category numbers of the parameters are regarded as individual dimension numbers d in a population in the optimization algorithm, each dimension respectively comprises the hidden layer node numbers, the time step lengths, the learning rates and the sub-module numbers, the constraint condition is that each parameter is positive, and the node numbers, the time step lengths and the sub-module component numbers are integers.
The parameter optimization process is as follows:
first, group initialization
In the d-dimensional function optimization problem, each solution in the feasible search space is an individual with a d-dimensional search vector. The upper and lower boundaries of each individual during the optimization are defined as UB and LB, respectively, and epsilon is a random number subject to uniform distribution over the interval 0, 1. Therefore, the initialization equation for the j-th dimension component of the ith individual in each population can be expressed as follows:
Pi(j)＝LB+ε×(UB-LB)
② fractal diffusion
Gaussian distribution is adopted as a random walk mode in the fractal diffusion process:
GW1＝Gaussian(μBP，σ)+(ε×BP-ε′×Pi)
GW2＝Gaussian(μP，σ)
in which epsilon and epsilon' are both intervals [0, 1]]Random numbers obeying uniform distribution; pi and BP represent the location and best individual, respectively, of individual i in the population; mu.sBP、μPAnd sigma are Gaussian parameters and respectively represent | BP |, | PiAnd standard deviation, where the standard deviation σ can be expressed as:
in the formula,
the step size for reducing the gaussian jump, where g represents the number of iterative computations.
③ updating Process I
In order to enable the individuals with better performance in the population of each generation to transmit individual information to the next generation with higher probability, firstly, calculating the fitness value of each individual in the population after initialization, sequencing according to the size of the fitness value, and defining the performance level for each individual:
wherein, rank (P)i) Representing the ranking of the ith individual in the population; n denotes the population size. For each individual P in the populationiIf the condition is determined
If yes, updating the j-th dimension component of the individual:
Pi′(j)＝Pr(j)-ε×(Pt(j)-Pi(j))
in the formula, PiIs' is PiAn updated position; prAnd PtIs an individual randomly selected from a population; ε is in the interval [0, 1]]Obeying uniformly distributed random numbers.
Update Process II
After the updating process I is executed, sequencing all the individuals in the population once again according to the sequencing method in the formula step III, and obtaining the performance level P 'of the sequenced individuals'giJudgment Condition P'giIf < ε is satisfied, then update the current individual according to:
In the formula, Pi' is the current individual; pi"is the updated individual; pt'and P'rRespectively randomly selecting individuals from the population after the first update;
is a random number generated by a Gaussian normal distribution
ε' is the following [0, 1] over the interval]Uniformly distributed random numbers.
Finally, carrying out fitness value comparison, setting a fitness function as the Mean Absolute Percentage Error (MAPE) of the model predicted value sequence and the actual time sequence, and calculating as follows:
wherein A represents the number of predicted results, siWhich represents the true result of the image,
indicating the prediction result. The average absolute percentage error represents the relative difference between the predicted result and the actual result, and the smaller the value, the better the predicted result.
When f (P) is satisfiedi″)＜f(Pi') when using the individual Pi"replacement of individual Pi', otherwise remain unchanged.
Finally, determining the stackThe generation condition G is more than GmaxWhether or not, wherein GmaxThe maximum number of iterations set for initialization is indicated, and g indicates the current number of iterations. If not, continuing to return to the step; otherwise, the algorithm is ended, a parameter result is output, the sub-mode number K in the result is used as the optimal sub-mode numerical value obtained by VMD decomposition in the step (II), and the number of nodes of the hidden layer, the time step length and the learning rate are used as the optimal parameters in LSTM optimization for calculation.
Fifthly, testing and calculating each sub-mode by using a test set, and performing index evaluation on errors; performing reverse normalization on the predicted value result meeting the verification condition, and outputting a predicted result;
and (4) substituting the test set T into the LSTM model with optimized parameters in the step (IV) for calculation to obtain a corresponding prediction data set F. The mean absolute percentage error of F and T was calculated. If the average absolute percentage meets the preset average absolute percentage, the LSTM model with the optimized parameters in the step (IV) is the final model;
inputting the value of the historical electricity consumption quantum modal component in the previous B months, the temperature, humidity and day type data processed by the nonlinear mapping of the B months into the LSTM model of the corresponding component obtained in the step (IV) to obtain the normalized electricity consumption predicted value of each component, then performing reverse normalization on the normalized electricity consumption predicted value of each component, and accumulating to obtain the electricity consumption predicted value Q in the current monthcapacity(ii) a When accumulating, accumulating each component inverse normalization value to obtain the predicted value of the monthly power consumption of the user; then accumulating the predicted values of the current month electricity consumption of all the users to obtain a predicted value Q of the current month electricity consumptioncapacityNamely, the retail capability is predicted for the power selling company in the current time period.
(VI) according to the predicted value Q of the electricity sold in the current monthcapacityAnd calculating the zero price difference delta p of the daily retail market batch of the power selling company.
The exchange calculates and calculates the daily retail market batch zero-price difference according to the latest daily wholesale transaction and retail transaction data
Wherein,
indicating the retail income of the electricity-selling company,
indicating the wholesale cost of the electricity selling company,
the calculation method is as follows:
wherein N isretailIndicating the number of retail customers of the electricity-selling company,
represents the retail market transaction price of the electricity consumer i,
representing the electricity consumption of the retail market of the electricity consumer i;
wherein Q iscapacityThe predicted value of the current month electricity consumption of the electricity selling company is shown,
showing the current month batch of the electricity selling companyThe electricity consumption is generated by the power generation device,
indicating the current monthly wholesale transaction average price of the electricity selling company, PdIndicating the monthly deviation electricity reference price, xioShowing the penalty factor xi of the electricity selling company for the excess electricityuPenalty coefficient, Q, indicating low power consumption for electricity selling companyoverThe calculation value of the potential electricity overuse amount of the electricity selling company is represented by the following calculation method:
Quthe calculation value of the potential power consumption shortage of the power selling company is represented by the following calculation method:
Qunderthe method represents a potential less-used assessment electric quantity measurement value of an electricity selling company, and the calculation method is as follows:
deviation electric quantity reference price PdThe calculation formula is as follows:
in the formula, QClean and superhairFor clean energy power plant monthly total clean super power generation quantity, QThermal power super power generationFor the monthly integral net excess power generation quantity, P, of the thermal power plantThermal power hanging plateUnified listing price, P, for regulating electric quantity for thermal power0And showing the up-regulation service reference price of the region.
Thus, stare-to-stare calculation is completed for retail market lot zero-order differences under the time scale of the daily cycle.
And (seventhly), tracking the retail transaction risk of the market based on the daily calculated retail market batch zero-valent difference calculation result delta P, and matching corresponding early warning levels and early warning measures, thereby completing the retail market risk tracking and early warning.
When daily risk tracking is carried out, if the daily retail market batch zero-price difference is less than 0 and the retail income loss reaches alpha of the performance guarantee fund amountrisk(αriskIf the current time is less than 1), triggering first-stage early warning, and sending an operation attention letter to the power selling company by the power grid company to remind the power selling company to pay attention to self settlement risk;
beta when the daily retail market batch zero-price difference is less than 0 and the retail income loss reaches the performance guarantee fund amountrisk(αrisk＜βriskIf the number of the power grids is less than 1), triggering second-stage early warning, and sending an operation inquiry letter to the power selling company by the power grid company;
if the daily retail market batch zero-price difference is less than 0 and the retail income loss reaches the gamma of the performance guarantee fund amountrisk(φrisk≤γrisk＜1Uφrisk＞βrisk) Triggering a third-level early warning, sending a risk warning letter to the power selling company by the power grid company, taking a settlement suspension measure, and implementing the settlement recovery according to the next day after the completion of the subsidy payment and performance guarantee deposit;
if the daily zero-price difference income trial value is less than 0 and the loss of retail income reaches gamma of the fulfillment guarantee fund amountrisk(ΓriskNot less than 1), triggering the fourth-level early warning, taking fusing measures by the power grid company, immediately taking pause settlement measures for the power selling company, and deducting the market credit point P of the power selling companyscoreWhen P isscore≥Nrisk(NriskRepresenting market credit point risk threshold set by the power grid company), performing corresponding market admission settlement period limiting measures on the power selling company, namely, settling the power grid company again after a time period delta psi after paying the performance guarantee bond. Wherein N isriskRepresenting a market credit point risk threshold, alpha, set by the grid companyrisk、βrisk、γrisk、ΓriskRespectively represents the proportion of the retail yield loss to the operating bond deposit phi when the first, second, third and fourth-level early warning are triggeredriskAnd the lower limit proportion of the retail yield loss to the performance bond guarantee fund is shown when the third-level early warning is triggered.
Examples of the applications
Referring to fig. 1, a method for monitoring, controlling, and warning risks in retail markets of power selling companies according to an embodiment of the present invention is shown. The power consumption prediction of a certain regional power selling company in Yunnan province is taken as an example. The temperature and humidity of the service area of the electricity selling company have large changes, the cold and the heat are distinct in all seasons of the year, the coldest is 1 month, the average temperature is 3 ℃, the most hot is 7 months, the average temperature is 20 ℃, the extremely high temperature value is 31 ℃, and the extremely low temperature value is-6 ℃. The month with the highest humidity in the region is 5 months, the average humidity is 80%, the month with the lowest humidity is 12 months, the average humidity is 62%, the extremely low humidity value is 56%, and the maximum humidity value is 87%.
Step 1: 365 data of electricity consumption of 1 year calendar history of a certain user corresponding to the electricity selling company are obtained. Meanwhile, the system automatically collects the temperature and humidity conditions and the day type data of the local city where the user is located, and stores the data into the database. And (4) performing stability test on the obtained historical power consumption data according to the obtained historical power consumption data, and selecting a KPSS test method to test the data stability. If the assumed result is rejected, i.e. the time sequence is not stable or the trend is not stable, the data is subjected to data cleaning, and abnormal values and missing values are processed firstly. And taking the abnormal value as a missing value, uniformly processing the two types of abnormal values, and interpolating the missing value by adopting a quadratic spline interpolation method. After the steps are completed, in order to eliminate the influence of unit inconsistency among different influence factors and accelerate the model processing efficiency, the data is normalized.
Step 2: and processing the power consumption sequence processed in the step one by utilizing a VMD (virtual machine tool) by adopting a non-recursive and variational modal decomposition method. Decomposing the power consumption time series f into K submodels, namely a finite bandwidth mode function { uk(t) }, K1, 2, 3, K, for each sub-modal function uk(t) calculating the corresponding analytic signal by using Hilbert transform:
for each mode function uk(t) center frequency ω of the FM signal by each single component amplitude modulation corresponding theretokIndex term of
Aliasing:
solving the signal bandwidth of the submode according to a Gaussian smoothing method for mediating signals, wherein an objective function is as follows:
wherein, { uk}＝{u1，...，uK}；{ωk}＝{ω1，...，ωK}。
And converting the problem into an unconstrained variational optimization problem by adopting a secondary penalty factor alpha and a Lagrange multiplication operator lambda (t):
pair sub-mode
Frequency of
And lagrange multiplier λn+1And performing alternate updating, wherein the final solution is as follows:
wherein,
respectively represent
f(t)、ui(t), λ (t), i ≠ k, n denotes the number of iterations.
Setting the precision epsilon of the termination condition of the iterative computationbCalculation accuracy εcThe following were used:
if epsilonc＜εbThe model stops iterating, otherwise, the loop calculation is continued.
And step 3: and constructing an influence factor mapping set for relevant influence factors influencing the electricity consumption of the user, wherein the relevant influence factors comprise temperature, humidity and day type data, and the mapped time series are obtained by clustering the original time series and then adopting a nonlinear mapping database.
Because the temperature, humidity and day type data have great influence on the electricity consumption of the user, for example, under general conditions, the electricity consumption in weekends and holidays is obviously smaller than the electricity consumption of working days, the electricity consumption in hot seasons or cold seasons with high temperature has great influence on the electricity consumption of the user, and the unit dimensions of all the influence factors are different, the influence factors are clustered firstly, and then the different influence factors are mapped to a certain interval by adopting a nonlinear mapping method. The correlation of the historical power usage with the above-described correlation impact factor is first calculated. The correlation calculation is judged according to the index value Spearman rank correlation coefficient, the coefficient larger than 0.3 can be regarded as correlation, and the three influence factors can be considered in a prediction model because the calculated temperature, humidity and day type are all larger than 0.3.
1) For the temperature:
the lower temperature threshold is set totThe upper limit temperature threshold is set to
Setting the temperature factor tnThe pre-mapping value is xnThe mapped value is ynWhen is coming into contact with
Then, a normalization method is adopted for processing; when in use
Or t <tAnd then, processing by adopting a mapping relation function:
firstly, the average value of the user electricity consumption corresponding to each temperature is counted in the obtained user historical electricity consumption data set, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the temperature-electricity consumption average value, the fitting goodness of three fitting curves and the pre-mapping value is respectively calculated, namely, the coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein x isiFor the value before the i-th mapping,
represents the mean of the values before the mapping,
represents the ith mapped value;
after the fitting relation curve is determined, inputting a pre-mapping value to calculate an application electric quantity value, and normalizing the value in a corresponding interval to obtain a mapped value;
2) for the humidity:
the lower limit humidity threshold is set asrThe upper limit humidity threshold is set to
Setting the humidity factor rnThe pre-mapping value is unMapped value is wnWhen is coming into contact with
When the method is used, a normalization method is adopted for processing, and when r is used, the method is used for processingn＜r0When is coming into contact with
Or r <rAnd then, processing by adopting a mapping relation function:
firstly, the average value of the power consumption of the user corresponding to each humidity in the obtained historical power consumption data set of the user is counted, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the humidity-power consumption average value, the fitting goodness of three fitting curves and a mapping pre-value is respectively calculated, namely, a coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein ui is the ith pre-map value,
represents the mean of the values before the mapping,
represents the ith mapped value;
after the fitting relation curve is determined, inputting a pre-mapping value to calculate an application electric quantity value, and normalizing the value in a corresponding interval to obtain a mapped value;
3) for day type data:
day type data fnThe pre-mapping value is hnThe mapped value is znAnd processing by adopting a mapping relation function:
firstly, the average value of the user power consumption corresponding to each day type is counted in the obtained user historical power consumption data set, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the day type-power consumption average value, the fitting goodness of three fitting curves and a mapping pre-value is respectively calculated, namely, a coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein h isiFor the value before the i-th mapping,
represents the mean of the values before the mapping,
representing the ith mapped value.
And 4, step 4: and carrying out data set division on the normalized user historical electricity consumption data set D in a cross validation mode. The first 60% of the data was taken as training set M and the last 40% as test set T. Wherein, MT test distribution is consistent with D, so as to improve the adaptability of the subsequent model. And (3) constructing a long-short term memory neural network model by taking each mapping relation time sequence data set in the training set M as a model input factor and taking model parameters under various different hyper-parameters as output factors.
Establishing an LSTM neural network model, and performing optimization solution on LSTM parameters by using a random fractal search algorithm SFS, wherein the parameters comprise hidden layer node number, time step length, learning rate and sub-module component number K, the category number of the parameters is regarded as an individual dimension number d in a population in the optimization algorithm, each dimension comprises the hidden layer node number, the time step length, the learning rate and the sub-module number respectively, and the constraint condition is that each parameter is positive number and the node number, the time step length and the sub-module component number are integers.
The parameter optimization process is as follows:
first, group initialization
In the d-dimensional function optimization problem, each solution in the feasible search space is an individual with a d-dimensional search vector. The upper and lower boundaries of each individual during the optimization are defined as UB and LB, respectively, and epsilon is a random number subject to uniform distribution over the interval 0, 1. Therefore, the initialization equation for the j-th dimension component of the ith individual in each population can be expressed as follows:
Pi(j)＝LB+ε×(UB-LB)
② fractal diffusion
Gaussian distribution is adopted as a random walk mode in the fractal diffusion process:
GW1＝Gaussian(μBP，σ)+(ε×BP-ε′×Pi)
GW2＝Gaussian(μP，σ)
in which epsilon and epsilon' are both the intervals [0, 1]]Random numbers obeying uniform distribution; piAnd BP represents the location and best individual, respectively, of individual i in the population; mu.sBP、μPAnd sigma are Gaussian parameters and respectively represent | BP |, | PiAnd standard deviation, where the standard deviation σ can be expressed as:
in the formula,
the step size for reducing the gaussian jump, where g represents the number of iterative computations.
③ updating Process I
In order to enable the individuals with better performance in the population of each generation to transmit individual information to the next generation with higher probability, firstly, calculating the fitness value of each individual in the population after initialization, sequencing according to the size of the fitness value, and defining the performance level for each individual:
wherein, rank (P)i) Representing the ranking of the ith individual in the population; n denotes the population size. For each individual P in the populationiIf the condition is determined
If yes, updating the j-th dimension component of the individual:
Pi′(j)＝Pr(j)-ε×(Pt(j)-Pi(j))
in the formula, PiIs' is PiAn updated position; prAnd PtIs an individual randomly selected from a population; c is in the interval [0, 1]]Obeying uniformly distributed random numbers.
Update Process II
After the updating process I is executed, sequencing all the individuals in the population once again according to the sequencing method in the formula step III, and obtaining the performance level P 'of the sequenced individuals'giJudgment Condition P'giIf < ε is satisfied, then update the current individual according to:
In the formula, Pi' is the current individual; pi"is the updated individual; pt'and P'rRespectively randomly selecting individuals from the population after the first update;
is a random number generated by a Gaussian normal distribution
ε' is the following [0, 1] over the interval]Uniformly distributed random numbers.
Finally, carrying out fitness value comparison, setting a fitness function as the Mean Absolute Percentage Error (MAPE) of the model prediction data sequence and the verification set time sequence, and calculating as follows:
wherein A represents the number of predicted results, siWhich represents the true result of the image,
indicating the prediction result. The average absolute percentage error represents the relative difference between the predicted result and the actual result, and the smaller the value, the better the predicted result.
When f (P) is satisfiedi″)＜f(Pi') when using the individual Pi"replacement of individual Pi', otherwise remain unchanged.
Finally, the iteration condition G > G is judgedmaxWhether or not, wherein GmaxThe number of iterations set for initialization is indicated, and g indicates the current number of iterations. If not, continuing to return to the step; otherwise, the algorithm is ended, the parameter result is output, and the result is obtainedAnd (5) taking the sub-modal number K as an optimal sub-modal number obtained by VMD decomposition in the step (II), and calculating by taking the number of hidden layer nodes, the time step length and the learning rate as optimal parameters in LSTM optimization.
And 5: and substituting the test set T into the LSTM model with the optimized parameters for calculation to obtain a corresponding prediction data set F. The mean absolute percentage error of F and T was calculated. If the average absolute percentage meets the preset average absolute percentage, the LSTM model with the optimized parameters in the step (IV) is the final model;
inputting the value of the historical electricity consumption quantum modal component in the previous B months, the temperature, humidity and day type data subjected to nonlinear mapping processing in the previous B months into the LSTM model of the corresponding component obtained in the step (IV) to obtain a normalized electricity consumption prediction value of each component, and then accumulating after inversely normalizing the normalized electricity consumption prediction values of each component; and finally, by analogy, circularly calculating the power consumption of other users proxied by the power selling company to obtain a corresponding prediction set meeting the conditions, and finally accumulating the power consumption prediction value sets of all the users to obtain the power consumption Q in the monthcapacityNamely, the retail capability is predicted for the power selling company in the current time period.
Step 6: calculating and calculating daily retail market batch zero-price difference according to the latest daily wholesale transaction and retail transaction data
Wherein,
indicating the retail income of the electricity-selling company,
indicating the wholesale cost of the electricity selling company,
the calculation method is as follows:
wherein N isretailIndicating the number of retail customers of the electricity-selling company,
represents the retail market transaction price of the electricity consumer i,
representing the electricity consumption of the retail market of the electricity consumer i;
wherein Q iscapacityThe predicted value of the current month electricity consumption of the electricity selling company is shown,
indicating that the electricity selling company wholesaled the electricity consumption in the current month,
indicating the current monthly wholesale transaction average price of the electricity selling company, PdIndicating the monthly deviation electricity reference price, xioShowing the penalty factor xi of the electricity selling company for the excess electricityuPenalty coefficient, Q, indicating low power consumption for electricity selling companyoverThe calculation value of the potential electricity overuse amount of the electricity selling company is represented by the following calculation method:
Quthe calculation value of the potential power consumption shortage of the power selling company is represented by the following calculation mode:
Qunderthe method is characterized in that a potential scarce assessment electric quantity measurement value of an electricity selling company is represented, and the calculation mode is as follows:
deviation electric quantity reference price PdThe calculation formula is as follows:
in the formula, QClean and superhairFor clean energy power plant monthly total clean super power generation quantity, QThermal power super power generationFor the monthly integral net excess power generation quantity, P, of the thermal power plantThermal power hanging plateFor regulating the electric quantity for thermal power, the price of the tag is unified, P0And showing the up-regulation service reference price of the region.
Thus, stare-to-stare calculation is completed for retail market lot zero-order differences under the time scale of the daily cycle.
And 7: and tracking the retail transaction risk of the market based on the next batch of zero-price-difference calculation results in the daily period, and automatically matching corresponding early warning levels and early warning measures, thereby completing the tracking and early warning of the retail market risk. When daily risk tracking is carried out, if the daily retail market batch zero-price difference is less than 0 and the retail income loss reaches alpha of the performance guarantee fund amountrisk(αriskIf the current time is less than 1), triggering first-stage early warning, and sending an operation attention letter to the power selling company by the power grid company to remind the power selling company to pay attention to self settlement risk; beta when the daily retail market batch zero-price difference is less than 0 and the retail income loss reaches the performance guarantee fund amountrisk(αrisk＜βriskLess than 1), triggering second-stage early warning, and using power grid company to sell powerThe server sends an operation inquiry letter; if the daily retail market batch zero-price difference is less than 0 and the retail income loss reaches the gamma of the performance guarantee fund amountrisk(φrisk≤γrisk＜1Uφrisk＞βrisk) Triggering a third-level early warning, sending a risk warning letter to the power selling company by the power grid company, taking a settlement suspension measure, and implementing the settlement recovery according to the next day after the completion of the subsidy payment and performance guarantee deposit; if the daily retail market batch zero-price difference is less than 0 and the retail income loss reaches the gamma of the performance bond guarantee fund amountrisk(ΓriskNot less than 1), triggering the fourth-level early warning, taking fusing measures by the power grid company, immediately taking pause settlement measures for the power selling company, and deducting the market credit point P of the power selling companyscoreWhen P isscore≥Nrisk(NriskRepresenting market credit point risk threshold set by the power grid company), corresponding market admission settlement period limiting measures are carried out on the power selling company, namely, a time interval period is carried out after the performance guarantee is paid
And then the settlement can be carried out with the power grid company again. Wherein N isriskRepresenting a market credit point risk threshold, alpha, set by the grid companyrisk、βrisk、γrisk、ΓriskRespectively represents the proportion (10%, 25%, 50%, 100%) of the loss of retail income in the performance bond guarantee fund when the first, second, third and fourth-level early warning are triggeredriskThe lower limit proportion (45%) of the performance bond guarantee fund is occupied by the loss of the retail yield when the third-level early warning is triggered.
In order to analyze retail risks of electricity selling companies in certain areas of Yunnan, historical electricity consumption data of retail users, which are proxied by the companies within the whole year period from 3 months in 2020 to 2 months in 2021, are selected as experimental data, and the time dimension is divided into the following parts: in spring (3-5 months), summer (6-8 months), autumn (9-11 months) and winter (12-2 months of the next year), a cross-validation method is adopted to select the first 60% of time sequences as a training set M and the last 40% of data as a test set T in each season. The total of 10 experiments were performed, and the error value was the average of the errors calculated for all experiments.
The results are shown in Table 1
TABLE 1
| Spring MAPE | Summer MAPE | Autumn MAPE | Winter MAPE | |
| VMD-LSTM-SFS | 0.874％ | 0.757％ | 0.792％ | 0.653％ |
| VMD-LSTM | 1.902％ | 1.885％ | 1.639％ | 1.587％ |
| LSTM | 2.479％ | 2.315％ | 2.641％ | 2.308％ |
As can be seen from table 1, it is,the prediction errors of the VMD-LSTM-SFS models provided by the invention are smaller than the experimental errors of the VMD-LSTM and LSTM of the control group, so that a better prediction effect can be obtained, the self electricity selling capacity predicted by an electricity selling company is improved, and the retail capacity prediction can be carried out. Therefore, the electricity selling data of a certain electricity selling company in 12 months of the year is predicted, and the electricity reference price P is deviated in the monthd0.24392 yuan/kilowatt hour, the predicted zero price difference income of the power selling company on a certain day is-53.4 ten thousand yuan, the amount of the performance bond deposit accepted and paid when the payment is carried out with the power grid company is 80 ten thousand yuan, and the third-level early warning is triggered when the income loss reaches more than 50% according to the specified day, so that the power grid company sends a risk warning letter to the power selling company, takes a measure of suspending the settlement with the power selling company, correspondingly deducts the corresponding loss amount deposit from the performance bond deposit of the power selling company, and can sign a settlement agreement with the power grid company again after the company finishes the payment. If the daily zero-price difference profit of the power selling company is calculated to be-10.1 ten thousand yuan according to the previous week-to-market and synchronization percentage increase prediction method, and the power selling company is operated according to the first-level early warning, the method can effectively reduce the main body operation risk of each market of the retail market under the same condition, effectively monitor the market risk of the power selling company, better guarantee the rights and interests of central counter parties of the power grid company, greatly reduce the settlement risk of the retail power market, remarkably optimize the market operation efficiency, powerfully improve the market operation guarantee capability, and promote the smooth and orderly development of multi-market main bodies in the retail power market.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (8)
1. A retail market risk monitoring, prevention and control early warning method for an electricity selling company is characterized by comprising the following steps:
the method comprises the following steps that (1) historical electricity consumption data of a user corresponding to an electricity selling company, and temperature, humidity and day type data of a place where the user is located are obtained;
step (2), arranging the historical electricity consumption data obtained in the step (1) according to time sequence to serve as a historical electricity consumption time sequence, carrying out normalization processing, and then decomposing the normalized historical electricity consumption time sequence into K sub-modal components by adopting a VMD method;
step (3), carrying out nonlinear mapping processing on the temperature, humidity and day type data obtained in the step (1);
step (4), taking the value of the normalized historical power consumption quantum modal component in the previous B months, the temperature, humidity and day type data subjected to the nonlinear mapping processing in the previous B months as the input of the LSTM model, taking the value of the normalized historical power consumption quantum modal component in the current month as the output of the LSTM model, and training the LSTM model of each sub-modal component to obtain the LSTM power consumption prediction model of each sub-modal component; b is more than or equal to 1;
and (5) inputting the value of the historical electricity quantum modal component in the previous B months, the temperature, humidity and day type data subjected to the nonlinear mapping processing in the previous B months into the LSTM electricity consumption prediction model of the corresponding component obtained in the step (4) to obtain a normalized electricity consumption prediction value of each component, performing inverse normalization on the normalized electricity consumption prediction values of each component, and accumulating to obtain a current month electricity consumption prediction value Qcapacity；
Step (6), according to the predicted value Q of the electricity sold in the monthcapacityCalculating zero price difference delta P of retail market batches of daily electricity selling companies;
and (7) tracking the retail transaction risk of the market based on the daily calculated retail market batch zero-valent difference calculation result delta P, and matching the corresponding early warning levels and early warning measures to finish the retail market risk tracking and early warning.
2. The retail market risk monitoring, controlling and early warning method for the electricity selling companies according to claim 1, characterized in that in the step (1), the obtained historical electricity consumption data is subjected to stability test, and if the assumed result is rejected, the data is subjected to data cleaning.
3. The retail market risk monitoring, controlling and early warning method for the electricity selling company according to claim 2, wherein the specific method for data cleaning is as follows: processing abnormal values and missing values; and taking the abnormal value as a missing value, uniformly processing the two types of abnormal values, and interpolating the missing value by adopting a quadratic spline interpolation method.
4. The retail market risk monitoring, controlling and early warning method for electricity selling companies according to claim 1, wherein in the step (1), the types of days include working days, saturdays, sundays and holidays in four seasons of spring, summer, autumn and winter.
5. The retail market risk monitoring, controlling and early warning method for the electricity selling companies according to claim 1, wherein in the step (3), the specific method for the nonlinear mapping processing of the temperature, humidity and day type data is as follows:
1) for the temperature:
the lower temperature threshold is set totThe upper limit temperature threshold is set to
Setting the temperature factor tnThe pre-mapping value is xnThe mapped value is ynWhen is coming into contact with
Then, a normalization method is adopted for processing; when in use
Or
When the temperature of the water is higher than the set temperature,and (3) processing by adopting a mapping relation function:
firstly, the average value of the user electricity consumption corresponding to each temperature is counted in the obtained user historical electricity consumption data set, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the temperature-electricity consumption average value, the fitting goodness of three fitting curves and the before-mapping value is respectively calculated, namely, the coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein x isiFor the value before the i-th mapping,
represents the mean of the values before the mapping,
represents the ith mapped value;
after the fitting relation curve is determined, inputting a pre-mapping value to calculate an application electric quantity value, and normalizing the value in a corresponding interval to obtain a mapped value;
2) for the humidity:
the lower limit humidity threshold is set asrThe upper limit humidity threshold is set to
Setting the humidity factor rnThe pre-mapping value is unMapped value is wnWhen is coming into contact with
When the method is used, a normalization method is adopted for processing, and when r is used, the method is used for processingn＜r0When is coming into contact with
Or r<rAnd then, processing by adopting a mapping relation function:
firstly, the average value of the power consumption of the user corresponding to each humidity in the obtained historical power consumption data set of the user is counted, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the humidity-power consumption average value, the fitting goodness of three fitting curves and a mapping pre-value is respectively calculated, namely, a coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein u isiFor the value before the i-th mapping,
represents the mean of the values before the mapping,
represents the ith mapped value;
after the fitting relation curve is determined, inputting a pre-mapping value to calculate an application electric quantity value, and normalizing the value in a corresponding interval to obtain a mapped value;
3) for day type data:
day type data fnThe pre-mapping value is hnThe mapped value is znAnd processing by adopting a mapping relation function:
firstly, the average value of the user power consumption corresponding to each day type is counted in the obtained user historical power consumption data set, then quadratic function fitting, cubic function fitting and bell-shaped function fitting are respectively adopted for the day type-power consumption average value, the fitting goodness of three fitting curves and a mapping pre-value is respectively calculated, namely, a coefficient is determined
Determining a mapping relation function by using the determined coefficient closest to 1 as an index, and determining the coefficient
The calculation is as follows:
wherein h isiFor the value before the i-th mapping,
represents the mean of the values before the mapping,
representing the ith mapped value.
6. The retail market risk monitoring, controlling and early warning method for the electricity selling company as claimed in claim 1, wherein in the step (4), the LSTM parameter is optimized and solved by using a random fractal search algorithm SFS, wherein the parameter includes the number of hidden layer nodes, the time step and the learning rate;
in the step (2), K in the VMD method is determined by adopting a random fractal search algorithm SFS.
7. The retail market risk monitoring, controlling and early warning method for the electricity selling company as claimed in claim 1, wherein in the step (6), the specific method for calculating the zero price difference of the retail market lot every day is as follows:
calculating and calculating daily retail market batch zero-price difference according to the latest daily wholesale transaction and retail transaction data
Wherein,
indicating the retail income of the electricity-selling company,
indicating the wholesale cost of the electricity selling company,
the calculation method is as follows:
wherein N isretailIndicating the number of retail customers of the electricity-selling company,
represents the retail market transaction price of the electricity consumer i,
representing the electricity consumption of the retail market of the electricity consumer i;
wherein Q iscapacityIndicating the predicted value of the electricity usage in the current month,
indicating that the electricity selling company wholesaled the electricity consumption in the current month,
indicating the current monthly wholesale transaction average price of the electricity selling company, PdIndicating the monthly deviation electricity reference price, xioShowing the penalty factor xi of the electricity selling company for the excess electricityuPenalty coefficient, Q, indicating low power consumption for electricity selling companyoverThe calculation value of the potential electricity overuse amount of the electricity selling company is represented by the following calculation modes:
Quthe calculation value of the potential power consumption shortage of the power selling company is represented by the following calculation method:
Qunderthe method represents a potential less-used assessment electric quantity measurement value of an electricity selling company, and the calculation method is as follows:
deviation electric quantity reference price PdThe calculation formula is as follows:
in the formula, QClean and superhairFor clean energy power plant monthly total clean super power generation quantity, QThermal power super power generationFor the monthly integral net excess power generation quantity, P, of the thermal power plantThermal power hanging plateFor regulating the electric quantity for thermal power, the price of the tag is unified, P0And showing the up-regulation service reference price of the region.
8. The retail market risk monitoring, controlling and early warning method for the electricity selling companies according to claim 1, wherein in the step (7), the retail market transaction risk is tracked according to the daily retail market batch zero-order difference calculation result, and the corresponding early warning level and early warning measure are matched, and the method comprises the following specific steps:
if the daily retail market batch zero-price difference is less than 0 and the loss of retail income reaches the performance guarantee amount alphariskIn which α isriskIf the current is less than 1, triggering first-level early warning, and sending an operation attention letter to the power selling company by the power grid company to remind the power selling company to pay attention to self settlement risk;
if the daily retail market batch zero-price difference is less than 0, and the loss of retail income reaches the beta of the performance guarantee fund amountriskIn which α isrisk＜βriskIf the number is less than 1, triggering second-stage early warning, and sending an operation inquiry letter to the power selling company by the power grid company;
if the daily retail market batch zero-price difference is less than 0, and the retail yield loss reaches the gamma of the performance guarantee fund amountriskIn which phi isrisk≤γrisk＜1∪φrisk＞βriskTriggering a third-level early warning, sending a risk warning letter to the power selling company by the power grid company, taking a settlement suspension measure, and implementing the settlement recovery according to the next day after the completion of the subsidy payment and performance guarantee;
if the daily retail market batch zero-price difference is less than 0, and the loss of retail income reaches gamma of the performance bond guarantee fund amountriskWherein r isriskAnd (3) triggering a fourth-level early warning, taking a fusing measure by a power grid company, immediately taking a settlement suspending measure for an electric selling company, and deducting a market credit point P of the electric selling companyscoreWhen P isscore≥NriskThen, the electricity selling company is carried outCorresponding market admission and settlement period limiting measures, i.e. time period after paying the performance guarantee money
The later process can settle accounts with the power grid company again; wherein N isriskRepresenting a market credit point risk threshold, alpha, set by the grid companyrisk、βrisk、γrisk、ΓriskRespectively represents the proportion of the retail yield loss to the performance bond guarantee fund when the first, second, third and fourth-level early warning are triggered, phiriskAnd the lower limit proportion of the retail yield loss to the performance bond guarantee fund is shown when the third-level early warning is triggered.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111648252.1A CN114444883A (en) | 2021-12-29 | 2021-12-29 | Retail market risk monitoring, prevention and control early warning method for electricity selling company |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111648252.1A CN114444883A (en) | 2021-12-29 | 2021-12-29 | Retail market risk monitoring, prevention and control early warning method for electricity selling company |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114444883A true CN114444883A (en) | 2022-05-06 |
Family
ID=81365028
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111648252.1A Pending CN114444883A (en) | 2021-12-29 | 2021-12-29 | Retail market risk monitoring, prevention and control early warning method for electricity selling company |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114444883A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115239208A (en) * | 2022-09-21 | 2022-10-25 | 广东电力交易中心有限责任公司 | Electricity selling company feature model analysis method and system based on electric power transaction data |
-
2021
- 2021-12-29 CN CN202111648252.1A patent/CN114444883A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115239208A (en) * | 2022-09-21 | 2022-10-25 | 广东电力交易中心有限责任公司 | Electricity selling company feature model analysis method and system based on electric power transaction data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6837949B2 (en) | Prediction system and method | |
| TW552524B (en) | Methods and systems for automated inferred valuation of credit scoring | |
| US7165043B2 (en) | Valuation prediction models in situations with missing inputs | |
| US8577791B2 (en) | System and computer program for modeling and pricing loan products | |
| JP6735219B2 (en) | Prediction system and prediction method | |
| AU2288101A (en) | Methods and systems for efficiently sampling portfolios for optimal underwriting | |
| AU2444101A (en) | Methods and systems for finding value and reducing risk | |
| EP1259892A2 (en) | Rapid valuation of portfolios of assets such as financial instruments | |
| CA2362441A1 (en) | Methods and systems for optimizing return and present value | |
| WO2001050315A2 (en) | Methods and apparatus for automated underwriting of segmentable portfolio assets | |
| US20020019790A1 (en) | Cross correlation tool for automated portfolio descriptive statistics | |
| WO2009065134A1 (en) | Method for deal-based pricing and estimation of deal winning probability using multiple prospective models | |
| CN110852519A (en) | Optimal profit method considering various types of loads for electricity selling companies | |
| CN108388974A (en) | Top-tier customer Optimum Identification Method and device based on random forest and decision tree | |
| WO2001050314A2 (en) | Methods and systems for quantifying cash flow recovery and risk | |
| WO2006105170A2 (en) | Systems and methods for determining cost of capital for an entity in a bottom-up, fully risk-based manner | |
| CN111798301A (en) | Resource transfer quota determining method, device and system | |
| Shi | Application of the model combining demand forecasting and inventory decision in feature based newsvendor problem | |
| CN114444883A (en) | Retail market risk monitoring, prevention and control early warning method for electricity selling company | |
| Mihai et al. | Use of load profile curves for the energy market | |
| Ibrahim et al. | LRFM model analysis for customer segmentation using K-means clustering | |
| Dong et al. | Government financing for clean technology development: Financial risk and social benefits | |
| Fan et al. | Digital Marketing Technology Based on Electricity User Portrait and K-means Clustering Algorithm | |
| CN114399096A (en) | Transaction electric quantity optimization method based on reinforcement learning algorithm | |
| CN118798872A (en) | Prepayment method for steam heating network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |