CN108491969B

CN108491969B - Big data-based space load prediction model construction method

Info

Publication number: CN108491969B
Application number: CN201810222378.4A
Authority: CN
Inventors: 宋晓凯; 王世谦; 耿翠英; 陈上吉; 王文豪; 林烽; 刘洋; 郭建宇; 张龙; 娄北
Original assignee: State Grid Corp of China SGCC; Economic and Technological Research Institute of State Grid Henan Electric Power Co Ltd
Current assignee: State Grid Corp of China SGCC; Economic and Technological Research Institute of State Grid Henan Electric Power Co Ltd
Priority date: 2018-03-16
Filing date: 2018-03-16
Publication date: 2021-12-24
Anticipated expiration: 2038-03-16
Also published as: CN108491969A

Abstract

The invention provides a big data-based space load prediction model construction method, which comprises the following steps: 1) establishing a spatial prediction database; 2) classifying and sorting the economic data and the power load data, and constructing a load total prediction model; 3) carrying out land property analysis work; 4) dividing land in detail, and setting parameter types required by space load prediction; 5) determining a load density index and a load concurrence rate required by load prediction; 6) calculating the load prediction value of each functional partition by a space load prediction method, and representing the space distribution of the load by combining with a ground planning graph; 7) and verifying the output results of the processes in the load total amount prediction stage and the space load distribution prediction stage. The invention improves the traditional prediction method mainly based on the statistics of the total load consumption, further provides an algorithm model based on space load prediction, and provides a reliable basis for further construction and transformation of the distribution network planning region.

Description

Big data-based space load prediction model construction method

Technical Field

The invention relates to the technical field of planning and construction of power distribution networks, in particular to a space load prediction model construction method based on big data.

Background

The purpose of the urban distribution network load prediction research is to better plan service for the urban distribution network. The urban power planning specification states that urban power grid load prediction is mainly divided into two parts: the method comprises the steps of firstly, forecasting the total planning load of urban electric power, including forecasting the total load and the total electric quantity of a planning area, also called total load forecasting; the second is power partition prediction, that is, prediction of spatial distribution of load, also called spatial load prediction. To accurately plan the distribution points of the substations and the line channels of the power grid in detail, the distribution of the load space of the substations and the line channels must be determined before the design, so that the obtained grid structure can be more reasonable.

In order to ensure the continuous and rapid growth of economy and the harmonious development of society in China, the scale of a power system must be continuously enlarged to ensure the continuous supply of power. Therefore, in order to ensure that the power grid structure is reliable and safe, has strong adaptability and can meet the requirements of economic development and domestic power utilization of cities and towns, a planning department carries out a large amount of investment and large-scale construction on the construction and the transformation of the power grids in cities and towns and rural areas. In this process, the most important task is load prediction.

The invention with patent number 201710240575.4 discloses a power load prediction method, which comprises the following steps: acquiring historical daily data to form an evaluation matrix Mnm, wherein mij represents an index value of a jth index of an ith evaluation object in the evaluation matrix Mnm; performing linear transformation normalization processing on the evaluation matrix Mnm to obtain a normalization matrix Snm; calculating the weight of the similar day characteristic of each index in the matrix Snm; calculating the association degree between the prediction day and the historical day; sequencing the evaluation objects from large to small according to the degree of association, and selecting the front W groups of evaluation objects as training samples of a prediction algorithm; improving the prediction algorithm by using a genetic algorithm, optimizing the weight and the threshold of the prediction algorithm, and calculating the optimal weight and threshold for training; and inputting the information of the predicted day into an optimized prediction algorithm, and performing inverse normalization on the output value to obtain a power load value of the predicted day. The invention can predict the power load of the next day according to the historical day related data.

The invention with patent number 201410767416.6 discloses a power load prediction method, which comprises the steps of firstly collecting and utilizing historical data and inputting influence factors; clustering the same type of moments by adopting a clustering analysis method, classifying the moments mainly according to meteorological characteristic factors, and establishing the relationship between the load prediction error value at the next moment and the actual load prediction error values at the previous ny moments, the load prediction values at the previous nu moments and the system influence factors at the previous ne moments; establishing a data model according to the following formula; establishing a pseudo partial derivative estimation criterion function; establishing a prediction control input criterion function according to the following formula; and repeating the steps until the predicted value of the required target time is obtained. The present invention does not relate to load characteristic indexes, and does not establish a relationship between the load characteristic indexes and the influence factors. Aiming at a multidimensional and multilevel power load prediction system, the method adopts model-free load prediction control of a data driving theory to obtain an optimal load prediction value; and the accuracy and the real-time performance of load prediction are greatly improved.

The accuracy of space power load prediction is directly related to the rationality and economy of investment and operation of a power distribution system, and is the basis of grid planning of a power distribution network. Only if the future load size and distribution in the power supply area of the urban power distribution network are accurately predicted, the capacity and site selection of the transformer substation can be accurately guided, the power supply grids can be accurately divided according to the load distribution, and decision variables such as the trend and model of a feeder line, the installation of switch equipment, the input time of the switch equipment and the like are reasonably planned.

Disclosure of Invention

The invention aims to provide a big data-based space load prediction model construction method, improve the traditional prediction method mainly based on the total load consumption statistics, further provide an algorithm model based on space load prediction, and provide a reliable basis for further construction and transformation of a power distribution network planning region.

In order to achieve the purpose, the invention adopts the technical scheme that: the method for constructing the space load prediction model based on the big data comprises the following steps:

1) establishing a spatial prediction database, wherein the spatial prediction database comprises an entry unit and an external interface unit, and collects spatial prediction data, including collecting economic data, planning data and power load data in a distribution network planning area;

2) classifying and sorting economic data and power load data in a spatial prediction database, selecting one or more load total prediction methods, constructing a load total prediction model, and outputting a load total prediction result;

3) establishing a space load distribution model, obtaining land utilization data, total planning area development data and each part development data by analyzing the planning data in the step 1), confirming load current situation data by combining the power load data arranged in the step 2), finally carrying out land property analysis work, and determining classification results and area parameters required by space load prediction;

4) based on the land property analysis and load classification result of the step 3), performing detailed land division, and setting parameter types required by space load prediction aiming at the land division and load classification result, wherein the parameter types comprise a load density index and a load concurrence rate;

5) determining a load density index and a load concurrence rate required by load prediction based on the data received and sorted in the step 3) according to the parameter types set in the step 4);

6) calculating the load predicted value of each functional partition by a space load prediction method and the area parameter, the load density index parameter and the load coincidence rate parameter output in the steps, and representing the space distribution of the load by combining a ground planning map;

7) selecting existing time nodes by using the existing power load data, and verifying the output results of the total load forecasting and spatial load distribution forecasting stage processes; verifying the load prediction result, confirming the validity of the prediction result and outputting the result; and (4) verifying the problems existing in the overall calculation process and parameter setting from the step 3) to the step 6) and forming feedback when the load prediction result is not verified.

Furthermore, the input unit comprises an upper computer and a human-computer interaction interface connected with the upper computer, the external interface unit is respectively connected with the electric power GIS database and the EMS external database, and the external interface unit acquires initialization data and realizes bidirectional interconnection with the upper computer.

Further, the load total prediction method comprises a data statistics type prediction method and a fuzzy control type prediction method, the data statistics type prediction method comprises a trend extrapolation method, a regression model method and a time sequence method, and the fuzzy control type prediction method comprises a grey theory prediction method, an artificial neural network method, a fuzzy prediction method, an expert system method and a preferred combination prediction method.

Further, the economic data comprises national economy and social development historical data of a power distribution network planning region and latest national economy and social development plans; the planning data comprises the overall planning of urban construction of the planning area of the power distribution network, the detailed planning of controllability of each district, the overall planning of land utilization, the planning of environmental protection and the planning of special electric power items; the power load data comprises the historical annual power consumption condition of the power distribution network planning region, the power supply scale, the load characteristic curve, the load distribution condition, the load structure and the load density of various users.

Further, the spatial load prediction method is to subdivide a power supply area into a plurality of unit partitions, aggregate the partitions with the same or similar load characteristics into classes, and then attribute the load spatial distribution prediction into the prediction of the load density of various load partitions, and the main method flow comprises the following steps: the detailed planning map of the control city → the partition → the collection of data → the calculation of the load density index of the partition → the calculation of the load distribution.

Further, the mathematical model of the space load prediction method is as follows: let ρ be_itThe predicted average density of the ith load in t years, (i is 1,2, …, m, and m is the total number of classifications), rho_t＝(ρ_1t,ρ_2t,…,ρ_mt) For classifying the load average density vector, the vector of the classification area in the partition j is S ═ S_1jt,S_2jt,…,S_mjt). Then, the load prediction value for partition j in t years can be expressed as:

in the formula eta_jtThe classification load in partition j is divided by the coefficient for t years.

Further, the land utilization data comprises a land utilization balance table, a land utilization planning map and a control range of the volume ratio of each unit partition, and the load current data comprises load distribution, a load structure and load density.

Further, the load density index comprises a building area load density index and a floor area load density index, and is obtained by adopting a reference experience database method and a load curve class deduction method.

Further, the load synchronization rate includes an intra-industry synchronization rate and an inter-industry synchronization rate, and the calculation formula is as follows: the load synchronization rate is (total load of industry 1 + total load of industry 2 + …) × inter-industry synchronization rate is [ (a (st) user load of industry 1 + b (th) user load of industry 1 + …) × inter-industry 1 synchronization rate + (a (nd) user load of industry 2 + b (th) user load of industry 2 + …) × inter-industry 2 synchronization rate + … ] × inter-industry synchronization rate.

The invention has the beneficial effects that:

1. when spatial prediction data of load prediction application is collected, the selection of the volume ratio is more accurate as much as possible on the basis of data, the respective development positioning and development gravity centers of different districts in a power distribution network planning region are considered, and the volume ratio selection range is preliminarily defined through the transverse comparison of the districts in a city; according to the existing data, the actual use of various plots in a certain plot is considered, the selection range of the plot volume ratio of each plot is reduced by the national standard guidance of the volume ratio, and finally the volume ratio selection value is determined.

2. When the spatial load distribution prediction is carried out, typical users with saturated or nearly saturated loads are selected in a distribution network planning area, the load curve is deeply researched, and the saturated load density index is obtained by analogy of the typical load curve. Load density indexes formed by the two methods are complementary, and a load curve method is taken as a main method and an experience database method is taken as an auxiliary method for a distribution network planning area with mature load development and high user load research feasibility; a power distribution network planning area with high load development speed and low user load research feasibility is mainly based on an experience database method and assisted by a load curve method. The two methods are divided into primary and secondary, and the secondary and the primary are used for determining the load density index parameters together.

3. According to the method, the ratio of the peak load after superposition to the sum of the peak loads before superposition can be obtained through superposition and normalization calculation of the typical load curves of the users in the distribution network planning area, and the ratio is taken as the coincidence rate. When the simultaneous rate is calculated, the load of the same type should pay attention to the sample universality and the ratio between the same type and different industries; the proportion of different land types in the distribution network planning area is considered for different types of loads, and the accuracy of simultaneous rate calculation is ensured.

4. The invention provides a load prediction result verification process for verifying the rationality of the load prediction result, and the prediction result is verified in the longitudinal and transverse directions. The longitudinal verification refers to verifying the load prediction result with the development of historical years in a planned region, and analyzing the difference between the prediction result and the historical year data; and the transverse verification means that the prediction result is compared with the prediction results in other areas with similar development positioning and comparable development levels or the current development stage, and whether the prediction results can be matched or not is analyzed. The overall process of the verification firstly considers background factors influencing the load development, then new indexes used in the verification are verified, a total amount prediction model and distribution prediction index parameters of space load prediction are verified, and finally a correction result of the verification is fed forward to a load prediction process to correct the load prediction result.

5. The space load prediction model established by the invention has some measures for improving the prediction fineness. The model is used for predicting the load of the distribution network planning area, the prediction result can effectively provide guidance for decision variables such as division of a power supply grid, planning of a distribution network target and a transition net rack, selection and input time of power grid equipment and the like, the model has positive effects on improving the practical efficiency of the distribution network, has strong effectiveness and practicability, and is worthy of wide popularization and use.

Drawings

FIG. 1 is a flow chart of the construction of the space load prediction model of the present invention.

FIG. 2 is a flow chart of the spatial load prediction model verification of the present invention.

FIG. 3 is a flow chart of the prediction of load density for load zoning in accordance with the present invention.

Detailed Description

Examples

As shown in fig. 1 to 3, the method for constructing the space load prediction model based on big data includes the following steps: 1) establishing a spatial prediction database, wherein the spatial prediction database comprises an entry unit and an external interface unit, and collects spatial prediction data, including collecting economic data, planning data and power load data in a distribution network planning area;

The basis of the method is that basic data are collected in the step 1), and economic data comprise national economy and social development historical data of a power distribution network planning region and latest national economy and social development plans; the planning data comprises the overall planning of urban construction of the planning area of the power distribution network, the detailed planning of controllability of each district, the overall planning of land utilization, the planning of environmental protection and the planning of special electric power items; the power load data comprises the historical annual power consumption condition of a power distribution network planning region, the power supply scale, a load characteristic curve, a load distribution condition, a load structure and the load density of various users. Meanwhile, the input unit comprises an upper computer and a human-computer interaction interface connected with the upper computer, the external interface unit is respectively connected with the electric power GIS database and the EMS external database, and the external interface unit acquires initialization data and realizes bidirectional interconnection with the upper computer.

The core of the invention is the selection of two methods, namely total load prediction and space load prediction, and the two methods are explained by combining specific operations as follows:

the load total prediction method comprises a data statistical prediction method and a fuzzy control type prediction method, wherein the data statistical prediction method comprises a trend extrapolation method, a regression model method and a time sequence method, and the fuzzy control type prediction method comprises a grey theoretical prediction method, an artificial neural network method, a fuzzy prediction method, an expert system method and an optimal combination prediction method.

The data statistical prediction method is generally based on historical data of the load or relevant influence factors, and predicts the future load by establishing a certain model, and mainly comprises a trend extrapolation method, a regression model method, a time series method and the like.

The trend extrapolation prediction method is to research the obvious change trend of the load under a certain condition, find out the development rule and predict the future load value according to the change trend. The method focuses on curve fitting extrapolation or recursive calculation extrapolation of the power load data sequence with obvious trend, and mainly comprises a moving average method, an exponential curve model, a smooth prediction method, a logistic model and the like. The method has the advantages that: less historical data and less work is required.

The regression model prediction method is to search the functional relation between the load and the influencing factors by statistically analyzing and processing the data of relevant factors such as economy, politics, environment and climate of the area where the load is located, so as to establish a regression model to predict the load. The regression model prediction method can be divided into unary regression and multiple regression according to the number of the independent variables; the fitting curves can be classified into linear regression and nonlinear regression. The method has the advantages that the model is based on the relation between the research load growth trend and other measurable influence factors, so the model parameter estimation technology is mature and reasonable, and the prediction process is simple. The method has the disadvantages that the method has higher selection requirement on the sample, and the sample has more stable development trend and better distribution rule; in addition, it is sometimes difficult to find a suitable regression equation.

The time series method is used for predicting the future load of a power distribution network planning region by searching the rule that historical load data change along with time and then establishing a time series model. The time series method can be classified into a random time series analysis method and a deterministic time series analysis method according to the difference of the processing methods. Common methods for determining time series analysis include Census-H decomposition and exponential smoothing, while common methods for randomness include state space methods and Box-Jenkins methods. The time series method is generally used for medium and long term power load prediction, and has the advantages that: the method is simple and convenient, has less workload, and can replace a regression analysis method under the conditions that related influence factors are more complex and data information cannot be obtained. The disadvantage is that it requires as much historical data as possible, limiting its application in small cities.

For the fuzzy control type prediction method, due to continuous development and wide application of various advanced theories and intelligent algorithms in recent years, the development of the prediction theory has a great breakthrough, and a plurality of new prediction methods appear at home and abroad, which are generally called as the fuzzy control type prediction method. Such methods are well suited for nonlinear, time-varying, multivariate, and uncertain power load prediction because they do not require prior knowledge of the relevant prior knowledge of the model structure and parameters, nor do they require complex system identification to build a mathematical model of the process. At present, a fuzzy control type prediction method has a good effect in the load prediction application of a power system, and can be mainly divided into a grey theory prediction method, an artificial neural network method, a fuzzy prediction method, an expert system method, an optimal combination prediction method and the like.

The grey system theory is firstly proposed by professor Duncong in China, and is about the theory of researching, solving and analyzing, modeling, predicting, deciding and controlling the grey system, the essence of the theory is that irregular original data are formed into a generation sequence with strong regularity again through accumulation generation to be modeled, data obtained by a generation model are generated through accumulation subtraction (inverse operation of accumulation generation) to obtain a reduction model, and the reduction model is just a prediction model. The grey prediction is to use grey model established by grey system theory to predict grey system, in the prediction process, the observed data sequence is not regarded as a random process, but regarded as a grey quantity or process with random time change, and the grey quantity is whitened gradually through accumulation or subtraction generation, so as to establish a model and make prediction. Since the power load meets the basic conditions of the gray prediction model, the gray theory can be applied to the power load prediction, and a GM (m, n) model (usually, a GM (1,1) model) is established to predict the load, which is a so-called gray model prediction method. The method has the advantages that: the method has the advantages of less required load data, suitability for analysis and prediction under the condition of poor information, simple principle, convenient operation, no need of considering distribution rules and variation trends, and suitability for medium-short term load prediction; the disadvantages are that: it only adapts to the load sequence with exponential growth trend, and when the data discrete degree (data gray scale) is larger, the prediction precision is lower.

The artificial neural network theory simulates the intelligent processing of human brain, and has self-adaptive function to a large number of non-structural and non-precise laws through information memory, autonomous learning, knowledge reasoning and optimized calculation. The artificial neural network has strong self-learning and complex nonlinear function fitting capability, is very suitable for power load prediction, particularly short-term load prediction, and is one of the practical prediction methods which are applied more at present. However, the number of layers of the model structure network and the number of neurons are mostly determined by experience, and there is no strict standard, and the model is likely to have better prediction accuracy in a certain system, but the prediction accuracy may be reduced greatly by replacing one system, so how to form a prediction sample and determine the number of neurons in the hidden layer of the network becomes the research focus of the artificial neural network prediction method. In addition, since the development and change of the medium and long term load is a non-stationary process (closely related to local political and economic policies), the acquisition of the medium and long term load history sample is necessarily difficult, and the obtained valuable sample is also necessarily limited, so that the application of the artificial neural network in medium and long term load total quantity prediction is limited.

The fuzzy theory is a theoretical method capable of effectively processing the problem that uncertain factors exist in engineering practice (such as difficulty in quantitatively describing related quantities and incomplete qualitative data), and the power load is influenced by a plurality of uncertain, important and irrespective related factors in actual development and change. Therefore, the fuzzy theory is applied to the load prediction of the power system, the problem of uncertain factors in the load prediction can be well solved, and the improvement of the load prediction precision is facilitated. The current fuzzy prediction method mainly comprises three methods: fuzzy clustering method, fuzzy similarity priority ratio method and fuzzy maximum closeness method.

The expert system is a computer software system built by a knowledge-based programming method, which possesses the knowledge and experience of experts in a certain field and can use the knowledge like the experts to make intelligent decisions through reasoning. A complete expert system consists of four parts, namely a knowledge base, an inference engine, a knowledge acquisition part and an interpretation interface.

In the medium and long term load prediction, the development of economy and power in each area has particularity, and a plurality of uncertain factors exist in the development of future loads, and if the development change of the future loads is not corrected by the experience and judgment capability of experts, the prediction precision is inevitably reduced, so the expert knowledge plays an important role in the medium and long term load prediction. It is the superiority of expert system prediction techniques over other methods. However, expert system prediction methods also have some limitations: the expert system has many factors to consider, and it is difficult to quantify various factors and their relationships, and thus it is difficult to reflect them exactly in the expert system. Processing the same question in the expert system may result in different results from expert to expert, so the expert system may respond differently to the same question. And the development and development period of the expert system is longer because the expert system needs longer time for spatial prediction data accumulation and model correction.

The preferred combinatorial prediction method has two types of concepts: the first method is a prediction method which selects proper weight to carry out weighted average on prediction results obtained by a plurality of prediction methods; and secondly, comparing the prediction models in a plurality of prediction methods, and selecting the prediction model with the best fitting goodness or the smallest error as the optimal model for prediction. The method integrates information contained in various single models, and optimal combination is carried out on the basis of maximum information utilization, so that the method can achieve the aim of improving the prediction result under most conditions.

Preferably, in the application of the combined prediction method in load prediction, the determination of the combined weight is key and is directly related to the precision of the final prediction result. The combined prediction method may be classified into a fixed weight combined prediction method and a variable weight combined prediction method according to a weight determination method. At present, most of domestic and foreign researches are variable weight combination prediction methods, which reflect the rules of load in development and change, accord with the load development trend and have high prediction precision.

The space load prediction method adopted by the invention is to predict the load density of classified load and then calculate the load value of each subarea according to the subarea area constitution. In the method, loads with the same property in all the partitions are processed in a unified mode, so that the loads of the same type in all the partitions are aggregated into a large class, and the load prediction of the partitions in the whole power supply area can be summarized into the prediction of the loads of the large class and the distribution of the loads. Therefore, the classification load density and the partition area constitute the prediction which is the core of the classification partition prediction of the method.

The space load prediction method has the advantages of simplicity, convenience, easy acquisition of basic data, easy adaptation to the change of urban planning schemes, strong flexibility and the like, and is particularly suitable for the condition of relatively determined land properties in China, so that the space load prediction method is widely applied to production practice. The load prediction is the basis and the premise of power distribution network planning, and only the future load value and the position distribution of the power distribution network planning area are accurately predicted to provide reliable basis for further planning, construction and transformation of the power distribution network. For the grid planning of the urban distribution network, accurate prediction of the space load is particularly important. The grid division is selected to the wiring mode, the planning of the target net rack is planned to the drawing of the transition net rack, and a plurality of links of the grid planning are all established on the basis of space load prediction.

The land use scheme in China is basically formulated by planning departments and is effectively executed and implemented, namely the land use property of the land is basically determined in the future, so that land use decision is not a prominent problem which troubles space load prediction. The space load prediction by using the space load prediction method is just the main basis for determining the land type by using the urban planning scheme of the government on the land. The basic principle is as follows: the power supply area is subdivided into a plurality of unit partitions, the partitions with the same or similar load characteristics are grouped into classes, and then the load spatial distribution prediction is summarized into the prediction of the load density of various load partitions. It can be carried out according to the following main steps: the detailed planning map of the control city → the partition → the collection of data → the calculation of the load density index of the partition → the calculation of the load distribution. The load density prediction of various types of loads and the determination of the area of each partition are the core of the space load prediction method.

The mathematical model of the space load prediction method is as follows:

let ρ be_itThe predicted average density of the ith load in t years, (i is 1,2, …, m, and m is the total number of classifications), rho_t＝(ρ_1t,ρ_2t,…,ρ_mt) For classifying the load average density vector, the vector of the classification area in the partition j is S ═ S_1jt,S_2jt,…,S_mjt). Then, the load prediction value for partition j in t years can be expressed as:

From the above, the most critical point of the method is to determine the load density of each partition in the planning year. However, the existing general method is to use the average density of the same classified load in the whole distribution network planning area as the load density of each area in the planning period, obviously, the method is unscientific and unreasonable, and because the load histories or the development stages are different, even the load densities of the same type of load are far apart, the load prediction of each area based on the uniform average density of the classified load inevitably causes non-negligible errors.

The key point of the model construction process of the invention lies in the analysis and type selection of various data types, the determination and application of parameters, the analysis operation in each step is combined, and the following aspects are explained by combining specific examples:

1. analyzing the land property:

the space load prediction method is used for space load prediction work, firstly, the land utilization conditions of the distribution network planning area need to be sorted, and corresponding load classification is carried out according to the land planning classification conditions in the area. The planning area of the power distribution network is required to have complete and detailed land planning so as to determine the main category of the land property of the planning area of the power distribution network and the planning and construction area of various lands.

Referring to urban construction land property classification in urban land classification and planning construction land standard (GB50137-2011), the land property of an urban distribution network planning area is generally classified into eight major categories: residential sites, public management and public service sites, commercial service facility sites, industrial sites, logistics storage sites, road and transportation facility sites, public facility sites, and green and square site. Of these eight major classes, thirty or more middle classes and tens of subclasses can also be distinguished. In the space load prediction work, when loads in a distribution network planning region are classified correspondingly based on land property classification, eight types of loads are divided corresponding to land property large types. However, because the different land property middle classes in the same land property large class often have great differences in the construction standards and load levels (for example, between a cultural venue and a hospital, between a fueling station and a hotel), the land property in the land planning map is further refined to the middle class in the city planning, so as to better guide the city construction. When load classification is performed correspondingly, in order to meet the accuracy requirement of load prediction, the difference of the electrical loads among different land characteristics is reflected, and the load classification is generally refined to the medium-class level of the land characteristics.

In consideration of the actual land utilization conditions of the distribution network planning area and the specific requirements of regional development, the load classification can be further refined or simplified in the actual operation process. Under the condition that some land property middle classes have the same or similar load level and load characteristics, partial land property middle classes can be merged and expressed; for the situation that a more definite plot nature subclass is given in some plot plans, partial plot nature subclasses can be split again.

2. Capacity rate selection based on controllability specification

The volume ratio is also called as building area wool density, which is the ratio of the total building area to the land area in a land block in a certain area, and is an important index for measuring the development and use strength of the construction land. For users on a particular plot, the floor area and volume fraction determine their final building area level, namely: the user building area is the user occupied area multiplied by the volume ratio.

The different land properties determine that the land blocks have different construction levels in all stages of the urban construction work. Therefore, in the urban planning work, the appropriate volume rate needs to be selected for the plots with various land characteristics by combining various factors such as the actual conditions of various industries, the urban development level, the functional area development positioning and the like. The work is carried out along with the detailed compilation of city controllability, each district and each land block in the city planning range are detailed, and the detailed work is used as a guide of the development intensity control level of each land block in the city construction work.

In the space load prediction work, generally, a moderate volume rate standard value is selected according to the land property of each type of land parcel by referring to the volume rate selection range given in the detailed controllability of the distribution network planning region, and the building area accounting of each land parcel in the region is carried out. For the situation that a plurality of subareas exist in a distribution network planning area and are respectively provided with a complete controllability specification, and the controllability specification gives out selection ranges of different volume rates, the building areas of all the internal blocks of the distribution network planning area are calculated piece by piece according to respective controllability specification volume rate requirements.

However, due to the fact that various urban development planning schemes are different, detailed control refinement degrees are different, planning and compiling schedules are asynchronous, and the like, a power distribution network planning region faced by space load prediction work often does not have accurate and complete volume fraction indexes, and the practical problems that the volume fraction range of similar plots in the power distribution network planning region is too large, partial plot volume fractions are lost and the like are often faced. Considering the direct correlation between the volume ratio and the building area, the volume ratio with high uncertainty will inevitably affect the building area accounting of the distribution network planning area, and further affect the accuracy of space load prediction. Therefore, it is necessary to provide a flow of refining and selecting the volume ratio to narrow the selection range of the volume ratio and reduce the influence of inaccurate volume ratio data on load prediction.

The volume ratio refining selection process comprises the following steps:

(1) the zoning is based on city general rules, respective development positioning and development center of gravity of different zones in the power distribution network planning zone are considered, and the overall volume rate reference ranges of multiple property land among various zones (such as a central urban zone, a new urban zone, a development zone and the like) with lost volume rate selection ranges are firstly defined from the overall view of the power distribution network planning zone through the transverse comparison of the zones in the city. When the plot division defines the volume rate, the empirical data of the same-level city planning is fully referred to, the specific factors of the existing construction level of the old city, the radiation effect of the industrial center, the policy guidance of the development area of the new city and the like are considered, and the simplification and homogenization of the volume rate distribution are avoided as much as possible.

(2) The land parcel research is based on partial control rules or other alternative planning data (developer volume ratio reported data and the like) of the existing power distribution network planning region, considers the actual use of various land parcels in a certain region, and reduces the selection range of the volume ratio of each land parcel by the national standard guidance of the volume ratio by taking the land parcel as a basic unit.

(3) The trade benchmarking is to consider the difference of the general development level of the referenced city and the prosperity degree of the trade benchmarking, and to perform the trade volume rate benchmarking, which helps to further narrow the volume rate selection range of a certain type of land in a certain type of land. When the volume rate is selected by the same industry benchmarking, rough benchmarking of the same large land or unequal benchmarking of users with larger scale difference is avoided by screening effectiveness of benchmarking users.

3. Land use division based on land parcel function partition

In order to make the load distribution result closer to the actual situation, the land use planning map is used to finely divide the land use of the distribution network planning region according to the land use property analysis result. Generally, when space load prediction is carried out, the planning region of the power distribution network is divided into three layers, namely a large region, a middle region and a partition, wherein the middle region is used for determining the total load and the distribution condition in a certain range and then carrying out targeted medium-voltage network frame planning; in the grid planning, the term "partition" refers to one or more users with the same electricity consumption type or a single land, such as a factory, a residential area, or a business area composed of several shops, and the like, and is divided into partitions to determine the minimum unit for collecting parameters such as area, load, and the like, and effectively represent the spatial distribution of the load. The large areas are generally divided according to administrative partitions, power supply administration partitions and natural barriers, so that partition loads can be counted, cross-area power supply can be avoided during site selection, and meanwhile, the power grid plan of each large area can be conveniently butted with the local government plan. The spatial load prediction method is most suitable for the spatial load prediction method in a land-based division manner using a functional partition as a basic unit because various types of loads are classified first and applicable data of partition loads need to be easily collected.

4. Spatial load prediction parameter setting

As described above, according to the city general planning and the detail control planning, the area input parameters of each functional land block in the distribution network planning area can be obtained in order, that is: the method comprises the following steps of firstly, representing the size of land utilized by a user and the occupied area occupying the total area proportion of a power distribution network planning area; and the second step is used for representing the actual power utilization area of the user and calculating the subsequent load of the user.

As can be known from the formula (1), the space load prediction work is calculated based on three types of parameters, namely an area, a load density index and a coincidence rate, so that the load density of a distribution network planning area needs to be calculated by using a space load density method, and a load density index input parameter and a load coincidence rate input parameter need to be determined. Which comprises the following steps:

(1) load density index: firstly, a building area load density index is used for measuring and calculating the total power utilization load in a user building; secondly, the floor area load density index is used for measuring and calculating the electric load mainly based on illumination in squares, greenbelts, roads, parking lots and the like.

(2) Load synchronization rate: the intra-industry synchronization rate is used for quickly calculating the maximum load superposition of different users in the same industry (same load classification), and the synchronization rate is higher because the users in the same industry have the same electricity utilization property and the similar electricity utilization time; and secondly, the inter-industry simultaneous rate is used for quickly calculating the maximum load superposition among different industries (different load classifications), and the simultaneous rate is lower because the power utilization characteristics of users in different industries have larger difference and the power utilization time is often crossed.

5. Load density index investigation, analysis and selection

When the load prediction is carried out by adopting a space method, the load density index is a very key factor, and whether the load density index is properly selected can directly influence the accuracy of a prediction result. For a power distribution network planning region with a more complex load type, a load density index required by prediction is generally obtained by a reference empirical database method and a load curve analogy method.

(1) Method for referencing experience database

When the domestic planning personnel use the space load method to predict the space load of the distribution network planning area, the method of referring to the experience database is mostly adopted.

Generally, after the first stage of land occupation property analysis work, the corresponding load classification of each land parcel of the distribution network planning area is completed. And classifying various plots of the distribution network planning region in an existing experience database based on the load classification result, selecting the same or similar classification in the database for each plot to be predicted, and taking the load density index as the load density index reference value of the plot to be predicted.

And then, by combining the actual situation of the distribution network planning area, firstly considering the development level and development target of cities in each region, secondly considering whether population density and industrial scale specialization phenomena caused by environment, policy, historical reasons and the like exist, and accordingly correcting the load density reference value to form the load density index selection value of various plots in the distribution network planning area.

(2) Load curve analogy method

The method for determining the load density index by using the load curve analogy method is also a commonly used method for solving the load density index by planners at home and abroad at present. The method collects and arranges the load historical data of the existing power distribution network in the planning area of the power distribution network to be predicted, then analogizes the load density index of each subarea in the area to be predicted by fitting a load curve, and is generally applied on the basis that the area has certain load historical data.

From the development track of cities in economically developed countries, the load of a city is increased rapidly in the early stage of construction under the drive of rapid economic development. However, after the city is developed to a certain stage, the urban load is slowly increased or even stops increasing under the influence of resource conditions such as urban population, land, environment and the like. At this time, the load is conventionally said to reach a saturated state. Theoretically, the development process of urban load can be represented by a sigmoid curve. The annual growth rate of the load is higher in the early stages of urban development, and gradually decreases as the economic development approaches maturity and is relatively stable as the urban development stages change and time passes. When the urban load reaches saturation or is basically saturated, the development of the urban economic society shows very obvious characteristics, and the analysis deduces the commonness of the urban economic society, so that the urban economic society is helpful for making a reasonable judgment on whether the urban load level reaches the saturation state or the distance between the urban load level and the saturation state.

Based on the development rule of the urban load S-shaped growth curve, the load density change in the process that the load of the urban distribution network planning area naturally grows and gradually reaches saturation can be analyzed, and the load density index required by load prediction is determined. For a non-S-shaped load increase curve, the historical annual load curve conforming to the natural increase rule can be analyzed through careful selection of the sections, and the load prediction change can be analogized in the same way. However, the method for analyzing the load curve is not completely suitable for the special load increasing process with multiple inflection points and complicated change reasons. Other objective factors such as the transformation of the urban development direction and the construction thought, the change of policy and regulation and the like have great influence on the accuracy of selecting the load density index.

The steps of determining the load density index using this method are as follows:

1) after the first-stage land use property analysis work is completed, the power distribution network planning area is divided into functional land blocks and corresponding loads are classified, a plurality of typical users are selected for each type of loads in the power distribution network planning area according to classification results to conduct research work. When the users select the power users, the power users which reach saturation or are close to saturation should be selected as much as possible.

2) During investigation, data of building areas of various load users and typical daily maximum loads in historical years (within 3-5 years) are collected, year-by-year load density of each user is calculated, and changes of the load density are sorted and counted to form a load density increase curve of each user along with changes of time.

3) And fitting the load density increasing curves of a plurality of investigation points of the same type of load to form a load density increasing curve corresponding to each type of land blocks one by one along with the time.

4) And analogizing the future growth track according to the change rule of the load density growth speed of each load density growth curve, and further calculating the load density indexes of various plots year by year in the planning year.

The load curve analogy method is used for space load prediction, and the following points need to be noted:

first, the fitting of the load curve is based on extensive and detailed research and funding work, typical users of each type of load need to select three to five or more, and the historical data of the load and area should be collected for three to five years or more. The number of investigated users or the history year range is too small, which reduces the diversity of data base and further affects the accuracy of load index.

Secondly, the investigated typical user should select the user whose current annual load meets the saturation requirement (the annual growth rate is lower than 2%) or is close to saturation as far as possible, and at the same time, should accurately judge the stage of the annual load growth of the typical user, so as to avoid the influence on the determination of the load density index caused by the poor load curve convergence of the load data due to the overlarge load growth trend or the mismatching of different load curves on the time axis.

Third, for industrial and commercial financial services industrial loads, different users in the same load class may still have widely different load levels and load characteristics. For the situation, when a typical user to be researched selects, the proportion among users in different industries is required to be noticed, so that the load curve after fitting can more accurately and completely represent the load density increasing process of various users in the distribution network planning region, and the load characteristic of the whole load middle class is prevented from being replaced by the load characteristic of the single user.

6. Load concurrency rate calculation and selection

When the load calculation from the bottom to the top in the self-partitioning mode is carried out, because the power utilization laws and characteristics of different users in different industries and even in the same industry are different, the superposition of the total load amount cannot be simply added for calculation, and the intra-industry load concurrence rate and the inter-industry load concurrence rate need to be considered. Namely:

load synchronization rate (industry 1 total load + industry 2 total load + …) × inter-industry synchronization rate

(1 st user load + …) × industry 1 coincidence rate + (2 st user load + 2 th user load + …) × industry 2 coincidence rate + …) × inter-industry coincidence rate.

Due to different life styles and industrial detailed classification engineering structures in different dimensional areas or different cities, certain difference exists in simultaneous rate selection. Based on the situation, an estimation method for selecting the coincidence rate is provided, and the method adopts a coincidence rate estimation mode similar to a 'normalization method' by combining the adjustment of urban positioning and the difference of load composition ratios of various industries in planning year based on the investigation of the current situation.

(1) Non-industrial per-industry user concurrency rate estimation

1) Typical daily load curves of various types of land subarea power supply distribution transformer loads in the distribution network planning area are researched and researched, and at least more than 10 power supply distribution transformers with land characteristics are selected, so that the phenomenon of 'being approximate to the whole' is prevented.

Obtaining the peak value m-m (m) of the load point of each research user according to the typical daily load curve of each research user₁、m₂、m₃、m₄、……)。

2) Overlapping the load data of the load curves of the same type of industries to obtain a load characteristic curve of a certain industry in the current year; and converting the superposed load peak value into 1 to obtain a load characteristic curve with the peak value of 1 (namely, a normalizing process), namely calculating the coincidence rate in each industry.

Superposing the load curves of the research points in the same industry to obtain a load peak value n in the same industry; the coincidence rate eta in each industry can be obtained_a ₁＝n₁/(m₁+m₂+m₃+m₄+……)。

(2) Estimation of the coincidence rate of various users in an industry

Industrial loads are accounted in the same way as other types of loads, but with additional consideration of the types of user jobs contained within the distribution grid plan.

1) Similar to the method of other types of load users, typical daily load curves of various industrial subarea power supply distribution transformers in the planning area of the power distribution network are researched and researched. The operation types of the industry (such as chemical industry, textile industry, smelting industry, mechanical processing industry or small manual industry) need to be fully considered when selecting the distribution and transformation point, and the load characteristic curve difference of different industries is also very large due to different working time of the work and different load conversion capacities of the newspaper and the package.

Load peaks of different detailed industrial classification investigation points can be obtained according to investigation results.

2) The load characteristics of different industries have larger difference, the proportion of different operation types (such as chemical industry, textile industry, smelting industry, mechanical processing industry or small handwork) in the industry in saturation year needs to be considered, and investigation points are selected according to the proportion of users with different operation types and power utilization rules in the industry; after existing distribution and transformation curves in the corresponding distribution network planning area are investigated and overlapped, normalization processing is carried out, and the concurrence rate in the industry is obtained.

Load ratios of different industrial types need to be estimated by combining distant view year saturated load prediction and regional positioning; and obtaining a load characteristic curve after industrial normalization through the load curve, and calculating the synchronous rate in the industry.

(3) Estimation of load concurrency rate among industries

And predicting the space load of each partition by a space load prediction method to obtain the load peak values of various types of partitions in the distribution network planning area.

The load peak of a plurality of subareas in saturation years is represented by A; eta_aRepresents the above estimated intra-industry concurrency rate; codes 1,2, 3, … … represent various industries, where let 1 represent business. There are commercial individual partition load peaks summed as a₁。

Calculating the load peak value N of each industry in the distribution network planning region of the saturated year, and calculating the load peak value N of the saturated year business₁＝A₁*η_a ₁。

Load peak N of saturation year by each industry₁，N₂，N₃… … the load ratios between the industries can be obtained.

According to the load proportion, carrying out 'normalization' treatment on the load curves of various industries; and multiplying the normalized load curve by the load peak value of each industry saturated year, and superposing the load characteristic curves of each industry to obtain the total saturated year value B of the distribution network planning area.

The coincidence rate eta between each industry_b＝B/(N₁+N₂+N₃+……)

Inter-industry simultaneity η_bAiming at a specific power distribution network planning block, different blocks with large difference are formed in industries, and the synchronization rate between the industries has large difference and needs to be considered according to the power distribution network planning blocks.

In the calculation process, the more investigation points are selected, the more the measured concurrency rate and the total load value tend to be practical; in general, the saturation load concurrency rate among industries in a power distribution network planning region is 0.7-0.8, and the more single industry, the higher the concurrency rate.

7. Load spatial distribution prediction

For any load partition i, the future load W of the partition can be obtained by multiplying the load density ρ of the partition by the area S of the partition_SThe predicted value of (c), namely:

W_Si＝ρ_i×S_i

and repeating the calculation steps for each partition in the distribution network planning area, and completely predicting the load value of each partition divided by various loads by using the divided partition area and the determined load density index.

For any load middle area j, accumulating the areas of all the subareas contained in the load middle area j, and considering the in-industry simultaneity rate eta_aSynchronization rate with industry eta_bThe future load W of the middle area can be obtained_MThe predicted value of (c), namely:

in the formula, i is a partition number, and l is an industry category number.

According to the calculation result, a target annual prediction load value list of each partition and each middle partition in the distribution network planning area can be calculated. The prediction result is combined with a map for the distribution network planning area, so that a visual and visible load prediction value space distribution map and a load density thematic map can be formed.

The spatial distribution map shows the horizontal distribution condition of the loads at each part of the distribution network planning region by indicating the load value for each load region; the load density theme graph represents the contrast condition of the load density of each part of the distribution network planning region by adding different color blocks for each load region.

8. Prediction result verification

Since the load prediction work is a work for reasonably estimating the future load development level with a view to the future, it is impossible to simply and definitely judge whether the predicted result is correct or not. The measurement of the accuracy of the prediction result is more to verify the rationality of the prediction result through comparison and mutual verification among related data, so as to confirm whether the prediction result is effective or not.

When the load prediction result is verified, the comparability of verification is emphasized, and the result with large difference of the area and the load characteristic is avoided from being adopted for analogy. The verification of the prediction result is generally divided into two aspects: firstly, longitudinal checking refers to checking a load prediction result with the development of a planned region in a historical year, and analyzing the difference between the prediction result and the historical year data, such as whether the growth rate accords with the development stage of the region, whether the annual maximum load utilization hours is consistent with the adjustment level of the regional industrial structure, and the like; and secondly, performing transverse verification, namely comparing the predicted result with the predicted results in other areas with similar development positioning and comparable development levels or the current development stage, and analyzing whether the predicted result can be matched with the predicted result or the current development stage.

(1) Comparison method adopted for verification

The method and indexes used for verifying the load prediction result are generally a classical prediction method and related indexes thereof for intuitively reflecting the comparison condition of the load prediction result on the basis of a simple and reliable principle. The specifically adopted comparison method comprises the following steps:

1) transverse comparison: according to the prediction result, indexes such as average load density, comprehensive electricity consumption per capita, life electricity consumption per capita and the like can be compared in an analogy mode with the same area, and the rationality of the prediction result is determined.

The average load density index is a ratio of the total load of the distribution network planning area to the floor area of the distribution network planning area and is used for expressing an average power utilization value of each unit land area. Because the social economy and the power load of the cities are always discontinuously (in a leap way) developed along with a certain factor, a plurality of cities with similar development positions and different development levels are developed, and the average load density has stronger comparable significance. Therefore, the average load density index is a relatively intuitive method for representing the urban load development level.

2) Longitudinal comparison: the electric elasticity coefficient and the yield value consumption method can be adopted for checking, and the rationality of the prediction result is verified through the annual development rule.

The electric power elasticity coefficient is a ratio of an electric power consumption increase speed to a national production total increase speed in a period of time, is used for evaluating a general relation between electric power and economic development, and is usually used for investigating the matching degree of urban or regional energy development and economic development. For the annual elastic coefficient of the same city, the value of the elastic coefficient generally keeps stable or has a stable change trend along with the adjustment of the urban industrial structure without drastic change.

The unit consumption of production value refers to the ratio of the electricity consumption of the first, second and third industries in a period of time to the total production value, and is used for measuring the electricity consumption consumed by each unit economic value. For annual production value and unit consumption of the same city, the numerical value of the urban.

(2) Check reverse-pushing correction process

When the associated data are found to be out of order in the comparison and verification, performing reverse-thrust verification according to the load prediction stage process and the complete sequence of verification, and considering other factors influencing the load development. The specific idea is as follows:

1) background factors affecting the load development were examined.

The prediction result of the urban load is widely connected with various factors such as urban construction planning, industrial composition, economic development speed and the like. When the load prediction result has certain possible problems, such as fluctuation of data compared with historical years, unbalanced load distribution and change of load structure, the load prediction parameter and the calculation result are not simply attributed to errors, and relevant background factors influencing the load development are firstly checked.

And (3) by utilizing the collected spatial prediction data, and reading the industrial development planning, the latest work report of the government and the land utilization condition, determining whether the characteristics shown by the load prediction result are matched with the strategic planning direction of the urban development. Through the inspection of the step, the problem is positioned in the index inspection stage.

2) And (5) checking the newly added indexes during verification.

Except for the average load density verification, when the verification is carried out through other comparison indexes, corresponding new indexes are added according to the selected method. For example, when the comprehensive power consumption of each person is verified, a target annual population scale prediction index and a maximum load utilization hour prediction index of a power distribution network planning area need to be considered; when the production value is checked, the total GDP amount and industrial proportion prediction indexes of a target year in a power distribution network planning area need to be considered. The economic and population-related new indexes are generally derived from files of urban construction planning, industrial development planning and the like of a power distribution network planning area.

When the load prediction result cannot meet the check standard, the accuracy and the reliability of the newly added index are checked firstly. Such as whether the newly added index caliber is matched with the distribution caliber of the distribution network planning area, whether the source of the newly added index is reliable, whether the prediction process of the newly added index is accurate, and the like. Through the inspection of the step, the problem is positioned at the stage of the load prediction result.

3) And (4) checking a total quantity prediction model and a distribution prediction index parameter of space load prediction.

When the predicted value exceeds the reasonable interval in the result verification process, whether the total amount prediction model and the distribution prediction index parameter applied in the space load prediction are selected improperly or not is considered. The method mainly considers multiple aspects such as whether a total quantity prediction method with low historical load data utilization rate is selected, whether line-by-line industrial load density indexes meet the expected development level of a power distribution network planning area, whether the rate indexes meet the actual conditions of the power distribution network planning area and the like.

In practical application, the level of selecting the load distribution prediction index parameters is usually reversely measured through an average load density index; reversely measuring the change trend of the total load by an electric quantity method and an elastic coefficient method through human-average integration; the structural change trend of the load of each industry is reversely measured by a everyone living electricity consumption method and a production value unit consumption method.

4) And (5) positively feeding the checked correction result to a load prediction process to correct the load prediction result.

Based on the above-mentioned inspection process, the newly added indexes in the verification process and the prediction model and parameter indexes selected in the space load prediction are checked one by one to ensure the reliability of the basic data used and the accuracy of the selected process, and the targeted correction is carried out. And then, calculating the corrected result again according to the load prediction flow to obtain a new load prediction result.

While the invention has been described in terms of what are presently considered to be the preferred embodiments, and not of limitation, those skilled in the art will recognize that many alternative embodiments may be implemented using the concepts of the present invention. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. The method for constructing the space load prediction model based on the big data is characterized by comprising the following steps of:

2. The big-data-based spatial load prediction model construction method according to claim 1, wherein: the logging-in unit comprises an upper computer and a human-computer interaction interface connected with the upper computer, the external interface unit is respectively connected with the electric power GIS database and the EMS external database, and the external interface unit acquires initialization data and realizes bidirectional interconnection with the upper computer.

3. The big-data-based spatial load prediction model construction method according to claim 1, wherein: the load total prediction method comprises a data statistical prediction method and a fuzzy control type prediction method, wherein the data statistical prediction method comprises a trend extrapolation method, a regression model method and a time sequence method, and the fuzzy control type prediction method comprises a grey theoretical prediction method, an artificial neural network method, a fuzzy prediction method, an expert system method and an optimal combination prediction method.

4. The big-data-based spatial load prediction model construction method according to claim 1, wherein: the economic data comprises national economy and social development historical data of a power distribution network planning region and latest national economy and social development plans; the planning data comprises the overall planning of urban construction of the planning area of the power distribution network, the detailed planning of controllability of each district, the overall planning of land utilization, the planning of environmental protection and the planning of special electric power items; the power load data comprises the historical annual power consumption condition of the power distribution network planning region, the power supply scale, the load characteristic curve, the load distribution condition, the load structure and the load density of various users.

5. The big-data-based spatial load prediction model construction method according to claim 1, wherein: the space load prediction method is to subdivide a power supply area into a plurality of unit partitions, aggregate the partitions with the same or similar load characteristics into classes, and then attribute the load space distribution prediction into the prediction of the load density of various load partitions, and the main method flow comprises the following steps: the detailed planning map of the control city → the partition → the collection of data → the calculation of the load density index of the partition → the calculation of the load distribution.

6. The big-data-based spatial load prediction model construction method according to claim 1, wherein: the mathematical model of the space load prediction method is as follows: let ρ be_itThe predicted average density of the ith load in t years, (i is 1,2, …, m, and m is the total number of classifications), rho_t＝(ρ_1t,ρ_2t,…,ρ_mt) For classifying the load average density vector, the vector of the classification area in the partition j is S ═ S_1jt,S_2jt,…,S_mjt) Then, the load prediction value for partition j for t year may be expressed as:

7. The big-data-based spatial load prediction model construction method according to claim 1, wherein:

the land utilization data comprises a land utilization balance table, a land planning map and a control range of the volume ratio of each unit partition, and the load current data comprises load distribution, a load structure and load density.

8. The big-data-based spatial load prediction model construction method according to claim 1, wherein:

the load density index comprises a building area load density index and a floor area load density index, and is obtained by adopting a reference experience database method and a load curve analogy method.

9. The big-data-based spatial load prediction model construction method according to claim 1, wherein:

the load synchronization rate comprises an intra-industry synchronization rate and an inter-industry synchronization rate, and the calculation formula is as follows: the load synchronization rate is (total load of industry 1 + total load of industry 2 + …) × inter-industry synchronization rate is [ (a (st) user load of industry 1 + b (th) user load of industry 1 + …) × inter-industry 1 synchronization rate + (a (nd) user load of industry 2 + b (th) user load of industry 2 + …) × inter-industry 2 synchronization rate + … ] × inter-industry synchronization rate.