Power service intelligent evaluation method and system for small and medium-sized enterprises

Abstract

The invention discloses a power service intelligent evaluation method and system for medium and small enterprises, and relates to the technical field of power service evaluation, and the method comprises the following steps: multi-source data collection and preprocessing, construction of a special evaluation index system for the medium and small enterprises, multi-dimensional evaluation based on a deep learning model, and evaluation result output and service adaptation. According to the method, firstly, an exclusive evaluation system containing four types of first-level indexes such as cost suitability and electricity utilization stability and 12 special second-level indexes is constructed, and the characteristics of small and medium-sized enterprises are accurately matched; secondly, intelligent preprocessing is conducted on 7 types of systems by means of RESTfu 1API butt joint, evaluation is conducted in combination with CNN, BERT, LSTM + random forest fusion models, the error rate is reduced to 0.2% from 0.5%, and manual work is replaced to improve efficiency; thirdly, a visual report is output through a three-layer two-link framework, an adaptive scheme is pushed, exploration resources are efficiently scheduled, and the progress is synchronized; finally, cost reduction and efficiency improvement of small and medium-sized enterprises are facilitated, resource configuration optimization of power grid enterprises is facilitated, and fine management is realized.

Description

Technical Field

[0001] This invention relates to the field of power service assessment technology, specifically to a smart power service assessment method and system for small and medium-sized enterprises. Background Technology

[0002] Against the backdrop of artificial intelligence deeply empowering the digital transformation of the power industry, SMEs, as a vital component of the national economy, exhibit typical characteristics in their electricity consumption: relatively small scale of electricity consumption, high cost sensitivity, frequent fluctuations in electricity load, and urgent need to respond to expansion demands. They have an urgent need for accurate, efficient, and cost-effective electricity services. However, the existing electricity service evaluation system has not built a dedicated intelligent evaluation mechanism for these characteristics of SMEs. This leads to a common core problem for SMEs in accessing electricity services: the lack of an intelligent evaluation system tailored to their specific needs causes evaluation results to be disconnected from their actual requirements, resulting in a chain reaction of problems such as low service efficiency, imbalanced resource allocation, and insufficient cost control.

[0003] Specifically, the core issue manifests itself in three interconnected aspects: First, the assessment dimensions are mismatched with the characteristics of SMEs. Existing assessment systems largely adopt the assessment dimensions of general electricity users (such as large industrial users and residential users), focusing on conventional indicators such as total electricity consumption and voltage levels. They fail to incorporate specific dimensions that SMEs prioritize, such as cost sensitivity (e.g., the proportion of connection costs to operating costs), load fluctuation adaptability (e.g., differences in electricity consumption during peak and off-peak seasons), and expansion potential (e.g., the probability of short-term capacity expansion). This results in assessments failing to accurately reflect the true electricity service needs of SMEs. Second, the assessment methods rely heavily on manual labor and have low levels of intelligence. Existing assessments often depend on business personnel manually collecting data such as SMEs' electricity consumption records and payment records, relying on experience to complete the assessment. This not only takes 15-20 minutes per assessment but is also prone to errors in data entry and logical inconsistencies (e.g., discrepancies between enterprise registration information and electricity consumption records), leading to biased results. Furthermore, there is a lack of dynamic analysis capabilities for historical electricity consumption data of SMEs, making it impossible to proactively predict their service needs. Third, the assessment system lacks specific adaptation capabilities. The data interfaces and functional modules of existing power service systems (such as marketing management systems and power grid management platforms) are not designed for SMEs. On the one hand, it is difficult to integrate the data of SMEs scattered across multiple systems (such as business registration information, production and operation overview, and electricity consumption data). On the other hand, it is unable to output visualized assessment reports and service suggestions adapted to SMEs. As a result, the assessment results are difficult to directly guide the implementation of services, ultimately causing delays in power service response for SMEs (such as long processing cycles for emergency capacity expansion applications) and high service costs (such as unnecessary equipment configuration). At the same time, it also makes it difficult for power grid companies to accurately allocate resources to SMEs (such as distribution area load dispatch), affecting the overall power grid operation efficiency.

[0004] In summary, the existing power service evaluation system, due to the lack of a dedicated intelligent evaluation system for SMEs, results in a closed loop of problems: inaccurate evaluation, inefficient service, and waste of resources. This severely restricts the improvement of the power service experience for SMEs and the refined management of SME services by power grid companies. Therefore, this paper proposes an intelligent power service evaluation method and system for SMEs to overcome the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a smart evaluation method and system for electricity services for small and medium-sized enterprises, so as to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, this invention provides a smart power service assessment method for small and medium-sized enterprises, comprising the following steps:

[0007] Step 1: Multi-source data collection and preprocessing. Through standardized data interfaces, connect with relevant business systems in the power industry, including marketing, power grid management, metering automation, engineering management, business information, service platforms, and human resources, to collect basic attributes, electricity consumption behavior, business expansion projects, service interactions, and exploration resource data of SMEs. After filling missing values, removing outliers, and verifying data consistency, construct an evaluation dataset through standardized processing.

[0008] Step 2: Construct a dedicated evaluation index system for SMEs. Based on the characteristics of SMEs, such as small electricity consumption, high cost sensitivity, and frequent load fluctuations, and in conjunction with electricity user service standards, establish four primary indicators, including cost adaptability, electricity consumption stability, engineering compliance, and service demand potential, as well as corresponding special secondary indicators that reflect the characteristics of SMEs.

[0009] Step 3: Multi-dimensional evaluation based on deep learning model. Convolutional neural network, pre-trained language model and engineering drawing analysis technology are used to extract time series, text and engineering features from the evaluation dataset. The features are input into LSTM and random forest fusion model. The load fluctuation dynamic weight mechanism of industry production cycle plus entropy weight method is combined to calculate the initial evaluation value. Then the initial evaluation value is calibrated by calling the evaluation benchmark data of small and medium-sized enterprises in the power industry.

[0010] Step 4: Output the assessment results and adapt the service. Generate a visual report containing load weight analysis and engineering rectification suggestions according to the power industry assessment report specifications. Based on the assessment results, push the appropriate electricity price package and equipment transformation plan, coordinate the dispatch of survey personnel and reserve grid resources, and synchronize the service progress to the power service platform accessible to small and medium-sized enterprises.

[0011] Furthermore, in step 1, the standardized data interface is a RESTful API interface; missing value filling is done using the average daily data of small and medium-sized enterprises of the same voltage level in the same industry over the past 3 years; outlier removal is based on the judgment that the data exceeds 3 times the standard deviation of the industry average over the past 3 years; data consistency verification is performed by recognizing business licenses and engineering drawings through Baidu Smart Cloud OCR engine, combined with Drools rule engine verification; and standardization processing uses a normalization formula to map the data to the [0,1] interval.

[0012] Furthermore, the specific secondary indicators in step 2 are as follows:

[0013] Cost adaptability: percentage of electricity connection cost, on-time electricity payment rate, and compatibility with electricity pricing packages;

[0014] Power stability category: including load fluctuation coefficient with dynamic weight correction, power outage impact sensitivity, and backup power demand probability;

[0015] Engineering compliance: compliance rate of engineering drawings, standardization of equipment installation, and matching degree of meter models;

[0016] Service demand potential category: including short-term capacity expansion probability with dynamic weight adjustment, new energy access demand, and exploration response demand.

[0017] Furthermore, in step 3, the convolutional neural network has a 3-layer structure, takes the daily load data of the past 30 days as input, and outputs 128-dimensional time-series features; the pre-trained language model is a BERT-base model fine-tuned based on 100,000 power service work orders, and outputs 512-dimensional text features; the engineering drawing parsing technology is TeighaSDK, which extracts equipment parameters through regular expressions and analyzes line topology based on the NetworkX library.

[0018] Furthermore, the implementation process of the load fluctuation dynamic weighting mechanism in step 3 is as follows:

[0019] Step 1: Call the industry production cycle query interface to obtain the peak and off-peak seasons of the industries to which SMEs belong, and build a mapping table containing industry codes and cycle periods;

[0020] The second step is to divide the nearly one year of load data into months, calculate the load entropy value of each month using a formula, and then calculate the basic weight of each month using a formula.

[0021] Step 3: Adjust the weighting of peak season months by multiplying by 1.2 and off-season months by multiplying by 0.8;

[0022] Step 4: Use the corrected weights as weighting coefficients for time series features and input them into the LSTM model.

[0023] Furthermore, in step 4, the visualization report is in PDF format, including indicator score radar chart, monthly load weight line chart, and screenshots of non-compliant projects; the survey personnel are dispatched according to skill tags → GIS straight-line distance ≤ 5km → work order saturation in ascending order; the power grid resource reservation is based on the short-term capacity expansion probability setting capacity and 7-day validity period; and the service progress is synchronized to the Southern Power Grid online platform.

[0024] The intelligent power service assessment system for SMEs comprises a multi-source data integration layer, an intelligent assessment modeling layer, and an assessment result application layer. The system is integrated with existing power business systems and the entire power service process for SMEs. The multi-source data integration layer connects to marketing management, power grid management, metering automation, engineering management, business information, China Southern Power Grid Online, and human resources systems via RESTful API interfaces to collect and preprocess multi-dimensional data. The intelligent assessment modeling layer includes a data preprocessing module, a feature engineering module, an LSTM+random forest fusion model integrating a load fluctuation dynamic weighting submodule, and a result calibration module. The assessment result application layer includes a report generation module, a service suggestion module, and a resource scheduling module.

[0025] Furthermore, the collected data includes the skill tags of the surveyors, their WGS84 coordinate system positions, work order saturation, and historical survey completion rates. The report generation module of the evaluation result application layer synchronizes the report to the Southern Power Grid online platform through the report push API. The resource scheduling module calls the survey dispatch business intelligent agent API to match surveyors and sends a POST request to the power grid management platform to reserve resources. The scheduling results are synchronized to the surveyors' email terminals and the contacts of small and medium-sized enterprises.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] I. Addressing the issue of assessment dimensions not aligning with the characteristics of SMEs, and improving assessment accuracy:

[0028] Construct a dedicated indicator system (including 12 specific secondary indicators) encompassing cost adaptability, power supply stability, engineering compliance, and service demand potential:

[0029] On the cost side (the proportion of electricity connection costs and the compatibility of electricity price packages), it directly matches the cost-sensitive needs of small and medium-sized enterprises and can push installment payment and other solutions in a targeted manner;

[0030] The load weight is dynamically adjusted by industry production cycle and entropy weight method (e.g., 0.18 for October peak season and 0.08 for February off-season in manufacturing) to make the load fluctuation coefficient assessment more in line with reality.

[0031] On the demand side (probability of short-term capacity expansion and demand for new energy access), the demand for capacity expansion can be predicted to avoid service delays.

[0032] II. Addressing the issues of reliance on manual assessments and low accuracy, thereby improving assessment efficiency and reliability:

[0033] Multi-source data intelligent preprocessing: By connecting to 7 types of systems through RESTful API, combined with missing data filling with the industry average, anomaly removal with 3 times the standard deviation, and consistency verification with OCR + rule engine, manual input errors are eliminated, and high-quality datasets are built.

[0034] Deep learning fusion evaluation: CNN is used to extract workload features, BERT to extract work order text features, and TeighaSDK to extract drawing features. These are then input into an LSTM+random forest model and calibrated against industry benchmarks, resulting in a reduced evaluation error rate and improved confidence.

[0035] It replaces manual assessment (which used to take 15-20 minutes per assessment), significantly improving efficiency.

[0036] III. Addressing the lack of specific system adaptation and difficulties in service implementation in the assessment system, and optimizing service and resource allocation:

[0037] Based on a three-tier, two-link architecture (multi-source integration - intelligent modeling - result application, linking the entire business system and service process):

[0038] Output a PDF visualization report (including indicator radar charts, rectification screenshots and supporting evidence), and simultaneously push out the adaptation plan;

[0039] Survey and dispatch are conducted by skill matching → within 5km → low saturation screening, and power grid resources are reserved according to the probability of capacity expansion to avoid resource waste;

[0040] Service progress is synchronized in real time to the Southern Power Grid online platform and enterprises, improving transparency.

[0041] IV. Achieving a win-win situation for SMEs and power grid companies:

[0042] Small and medium-sized enterprises: reduce electricity costs (avoid inefficient configuration), shorten service response time (such as emergency capacity expansion and speed enhancement), and improve service experience;

[0043] Power grid companies: Optimize resource allocation (more precise load dispatching in distribution areas), reduce operating costs (replace manual assessment), and break the closed loop of inaccurate assessment - inefficient service - resource waste. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the intelligent power service assessment method and system for small and medium-sized enterprises of the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Please see Figure 1 The present invention provides a technical solution:

[0047] See Figure 1 The following is an example of an intelligent assessment method and system for electricity services for small and medium-sized enterprises:

[0048] I. System Architecture:

[0049] This system adopts a three-layer, two-link architecture. The three layers refer to the multi-source data integration layer, the intelligent assessment and modeling layer, and the assessment result application layer. The two links refer to the linkage with existing power business systems and the linkage with the entire process of power services for small and medium-sized enterprises. The technical implementation of each layer and module is based on the "Power Industry Data Exchange Interface Specification".

[0050] The specific structure and details of industry standards such as (DL / T1970-2018) and "Typical Design of 10kV and Below Business Expansion and Receiving Projects of Southern Power Grid" (hereinafter referred to as Southern Power Grid Typical Design) are as follows:

[0051] 1. Multi-source data integration layer:

[0052] This layer achieves data interoperability with the marketing management system, power grid management platform, metering automation system, engineering management system, business information system, and Southern Power Grid online platform through standardized RESTful API interfaces. It collects comprehensive data required for power service assessment of SMEs. The technical parameters, format, and frequency of data collection all comply with power industry data standards, as detailed below:

[0053] Basic attribute data: The system retrieves the electricity user profile query API (interface address: http: / / xxx.xxx.xxx / api / marketing / customer, request method: GET, timeout: 30 seconds) from the marketing management system to obtain the SME's electricity user ID, electricity type (low voltage / high voltage), industry classification (e.g., manufacturing C31, service industry O81), power supply unit, and voltage level. It also retrieves the SME's registered capital, years of operation, main business scope, legal representative's name, and ID number from the business registration information system via the enterprise information query interface (connected to the National Enterprise Credit Information Publicity System API, data format: JSON). This data is used to construct a basic user profile for SMEs, with a data update frequency of T+1 (synchronizing the previous day's data every morning).

[0054] Electricity consumption behavior data: The load data query API is called from the metering automation system to obtain the daily / monthly electricity load curves, peak-valley electricity consumption ratios (peak period 8:00-22:00, valley period 22:00-next day 8:00) and historical capacity increase / decrease records for SMEs over the past 1-3 years; the transformer area resource status API is called from the power grid management platform to obtain the line load rate (current load / rated load) and transformer capacity margin (rated capacity - current total load) for the transformer area where SMEs are located. The data collection frequency is real-time + T+1 (load data is collected in real-time, transformer area resource data is updated T+1); additionally, the industry production cycle query API is called (interface address:

[0055] http: / / xxx.xxx.xxx / api / marketing / industry-cycle (data source: National Bureau of Statistics Industry Production Cycle Database) obtains the peak / off-peak seasons for the industries to which SMEs belong (e.g., for manufacturing C31, the peak season is October-December and the off-peak season is February-March), which is used for subsequent dynamic load weight calculations.

[0056] Business Expansion Project Data: The system retrieves business expansion project data from SMEs over the past three years by calling the business expansion project archive API (interface address: http: / / xxx.xxx.xxx / api / engineering / archive, supports DWG format file download). This includes construction design drawings (marking equipment models and line cross-section specifications) and equipment lists (e.g., transformer model SCB14-800 / 10, line cross-section 120mm²). 2 ), completion acceptance records (including rectification items and re-inspection results), the data must comply with Chapter 3.2 of the Southern Power Grid Design Standard: Equipment Selection Specifications, and the update frequency is synchronized within 1 hour after the project is archived.

[0057] Service interaction data: The service work order API (interface address: http: / / xxx.xxx.xxx / api / csgonline / workorder) is called from the Southern Power Grid online platform to obtain service work order records of small and medium-sized enterprises (consultation type such as capacity expansion consultation, complaint content such as electricity bill questions, service response time) and online business processing records (installation progress, contract signing status). The data collection frequency is real-time (synchronized within 5 minutes after the work order is generated).

[0058] Exploration resource data: The system calls the Exploration Personnel Status API (interface address: http: / / xxx.xxx.xxx / api / hr / surveyor) from the Human Resources system to obtain information on exploration personnel within the jurisdiction, including skill tags (such as low-voltage engineering, high-voltage expansion), current location (latitude and longitude based on the WGS84 coordinate system of the GIS system), work order saturation (current number of work orders / maximum number of work orders that can be carried out, with the maximum carrying capacity set at 8 work orders per person per day), and historical exploration completion rate (work orders completed / work orders received in the last 30 days). The data is updated every 5 minutes and is used to evaluate service response capabilities.

[0059] 2. Intelligent evaluation modeling layer:

[0060] This layer constructs a multi-dimensional intelligent evaluation model based on deep learning algorithms and power industry business rules. The technical implementation details, algorithm parameters, and selection criteria for each module are as follows:

[0061] 2.1 Data Preprocessing Module: Missing value filling adopts the industry average filling method. For example, if a small and medium-sized enterprise is missing 3 days of load data, the average daily load of the same industry (e.g., manufacturing C31) at the same voltage level (10kV) in the past 3 years will be used to fill the missing values. Outlier removal is carried out according to the 3σ principle (outlier judgment standard: |x-μ|>3σ, where μ is the average load of the industry in the past 3 years and σ is the standard deviation). Extreme electricity consumption data caused by equipment failure (e.g., single-day load exceeds the average by 3 times) will be removed. Data consistency verification uses Baidu Smart Cloud OCR engine to recognize text information in business licenses and engineering drawings. The consistency between the electricity user name and the business name is verified by the rule engine (developed based on Drools rule engine) (using name + ID number double verification, matching threshold ≥95% is judged as consistent) and the compliance of the engineering drawing version (must be the Southern Power Grid Design 2023 version or later, judged by the drawing header annotation). Invalid data is automatically marked and removed.

[0062] 2.2 Feature Engineering Module: Electricity load time-series feature extraction uses a 3-layer convolutional neural network (CNN) with a 3×3 kernel, ReLU activation function, and max pooling (2×2 kernel). The input is daily load data from the past 30 days (720 dimensions, 15 minutes / time × 96 times / day), and the output is a 128-dimensional feature vector including peak-valley time variation trends and load growth slope. Service order text feature extraction uses a BERT-base model (fine-tuned based on electricity service order text, with a training dataset containing 100,000 historical service orders). Tokenization converts the text into a 512-dimensional vector, and attention mechanisms are used to extract keyword features such as emergency capacity expansion and excessively high electricity costs. Business expansion project drawing feature extraction uses TeighaSDK to parse DWG files, extracting equipment parameter features through regular expressions (e.g., transformer model: SCB14-\d+ / \d+), and extracting line topology features (e.g., power access path, equipment connection relationships) through graph structure analysis (based on the NetworkX library).

[0063] 2.3 Multi-dimensional evaluation module:

[0064] The system constructs four primary indicators—cost adaptability, power supply stability, engineering compliance, and service demand potential—and 12 secondary indicators (see Step 2 of the Intelligent Assessment Method in Example 2). A quantitative assessment is achieved using an LSTM + Random Forest fusion model, where the LSTM model incorporates a dynamic weight assessment submodule for load fluctuations, as detailed below:

[0065] 2.3.1 Dynamic Weighting Submodule for Load Fluctuations: Addressing the frequent load fluctuations characteristic of small and medium-sized enterprises (SMEs), this module breaks away from the existing fixed-weighting evaluation model and dynamically adjusts load data weights by combining industry production cycle data with the entropy weighting method.

[0066] Step 1: Industry cycle mapping. Call the industry production cycle query API from the multi-source data integration layer to obtain the peak / off-season periods of the industry to which SMEs belong (e.g., peak season for manufacturing C31 is October-December, off-season is February-March), and build an industry-cycle mapping table (stored in the marketing management system database, with fields including industry code, peak season start month, peak season end month, off-season start month, and off-season end month).

[0067] Step 2: Entropy weight calculation. Divide the nearly one year of load data into months and calculate the entropy value of the load data for each month (entropy formula:

[0068]

[0069] Where p ij The entropy value represents the proportion of the load on day i of month j to the total load for that month, where n is the number of days in the month. A smaller entropy value indicates a higher load.

[0070] This indicates that the greater the impact of the monthly load fluctuation on the assessment results, the greater the weight (weighting formula:

[0071]

[0072] Step 3: Cycle weight adjustment. Based on the industry-cycle mapping table, the weight of peak season months is multiplied by a correction factor of 1.2, and the weight of off-season months is multiplied by a correction factor of 0.8 (the correction factor is set based on the load assessment correction rules in Chapter 5.2 of the Southern Power Grid Design) to ensure that peak season load data (reflecting the peak production demand of small and medium-sized enterprises) has a higher proportion in the assessment.

[0073] Step 4: Integrate the weights into the LSTM, and use the corrected monthly weights as the weighting coefficients of the load time series feature vector (e.g., October weight is 0.15, and the corresponding load data is multiplied by 0.15 before being input into the model) to optimize the feature weight allocation of the LSTM model input layer.

[0074] 2.3.2 LSTM Model Main Structure: Used for power stability index assessment. The input is a 128-dimensional load feature vector after dynamic weight correction. The model structure is: input layer (128-dimensional) - hidden layer (128 units) - dropout layer (dropout rate = 0.2) - output layer (2-dimensional, corresponding to load fluctuation coefficient and short-term capacity expansion probability). The optimizer is Adam (learning rate 0.001), the number of iterations is 50, and the loss function is mean squared error (MSE). Through dynamic weight adjustment, the load fluctuation coefficient prediction is made to better fit the production cycle characteristics of small and medium-sized enterprises.

[0075] 2.3.3 Random Forest Model: Used for cost adaptability, engineering compliance, and service demand potential index evaluation. The input is cost data (electricity connection cost ratio, etc.), feature vectors of engineering drawings, and feature vectors of work order text. The model parameters are 100 decision trees, maximum depth of 10, Gini coefficient for feature selection, minimum number of sample splits of 5, and output is a three-level evaluation result of high / medium / low (e.g., high electricity connection cost adaptability).

[0076] 2.4 Result Calibration Module: This module calls the SME evaluation benchmark library API of the Marketing Management System (interface address: http: / / xxx.xxx.xxx / api / marketing / benchmark) to obtain the evaluation benchmark values ​​of SMEs in the same industry and with the same voltage level. It then uses a deviation correction formula (calibrated value = initial evaluation value × (benchmark value / industry average)) to calibrate the evaluation results, ensuring that the evaluation error rate is ≤0.2% (the accuracy is improved from the original 0.5% due to dynamic load weight adjustment), meeting the parameter verification accuracy requirements of Chapter 5.3 of the Southern Power Grid Design Standard.

[0077] 3. Evaluation Result Application Layer:

[0078] This layer enables the visualization of evaluation results and facilitates service implementation. The technical implementations of each module are integrated with existing power business processes, as detailed below:

[0079] Report generation module: Generates a PDF visual report in accordance with the "Electric Industry Assessment Report Standard" (DL / T2445-2021), including a cover (project name, work order number, assessment date), table of contents, and main text. The power consumption stability index page introduces a load fluctuation weight analysis chart, showing the dynamic weight distribution for each month (e.g., manufacturing industry weight 0.18 in October and 0.08 in February), and annotates the basis for weight correction (Chapter 5.2 of China Southern Power Grid Design Standard); the engineering rectification page includes screenshots of non-compliant drawings and the basis for rectification; the service suggestion page recommends suitable electricity price packages and equipment modification plans. After the report is generated, it is synchronized to the China Southern Power Grid online platform via the report push API.

[0080] Service suggestion module: Generates structured service solutions based on evaluation results and pushes them to the marketing management system through the service suggestion API. For example, for small and medium-sized enterprises whose electricity connection costs account for more than 5%, a phased payment plan for engineering fees is recommended (according to Article 4.2 of the "Southern Power Grid Business Expansion Project Fee Management Measures"); for enterprises whose meter model matching degree is less than 80%, it is recommended to adapt the metering equipment.

[0081] Resource scheduling module: The exploration personnel matching calls the exploration dispatch business intelligent agent personnel matching API. The algorithm logic is: skill tag matching → GIS straight-line distance (≤5km) → work order saturation sorting in ascending order → selecting the personnel with the lowest saturation. The matching results are pushed to the exploration personnel's Elink terminal. The power grid resource reservation sends a POST request to the power grid management platform, carrying the work order number, reserved capacity (20kVA is reserved for enterprises with a short-term capacity increase probability ≥60%), and a validity period of 7 days. After the platform returns a reservation success receipt, the work order status is updated.

[0082] II. Intelligent Assessment Method Steps:

[0083] This method, based on the above system architecture, is executed in four steps: data collection, indicator construction, model evaluation, and result application. The technical details and operational examples for each step are as follows:

[0084] Step 1: Multi-source data acquisition and preprocessing:

[0085] 1.1 Data Acquisition:

[0086] Through standardized API interfaces in the multi-source data integration layer, basic attribute data, electricity consumption behavior data, business expansion project data, service interaction data, and exploration resource data of SMEs are automatically collected.

[0087] Call the marketing management system's electricity user file query API, input the parameter user number: SG2024001, and get the returned data: electricity type: high voltage, industry classification: manufacturing C31, voltage level: 10kV, power supply unit: Guangzhou Power Supply Bureau Tianhe Branch;

[0088] Call the load data query API of the metering automation system, input the parameters user ID: SG2024001, time range: 2024-01-01 to 2024-12-31, and obtain the daily load data for the past year (e.g., maximum load of 120kW on 2024-10-15, maximum load of 40kW on 2024-02-10);

[0089] Call the marketing management system's industry production cycle query API, input the industry code: C31, and get the returned data: peak season: October-December, off-season: February-March;

[0090] Call the project expansion file API of the project management system, enter the work order number: GY2024005, and download the DWG format construction drawings (marking the transformer model SCB14-800 / 10).

[0091] 1.2 Data Preprocessing:

[0092] Missing value imputation: Load data for a certain small and medium-sized enterprise from May 1, 2024 to May 3, 2024 is missing. It is filled with the average load of 65kW of 10kV enterprises in the same industry (manufacturing C31) from May 1 to 3 over the past 3 years.

[0093] Outlier removal: On July 15, 2024, a company's load data was 300kW (more than 3 times the average of 90kW over the past 3 years), which was identified as an outlier. After removal, it was replaced with the average load of 88kW for the 3 days before and after that date.

[0094] Data consistency verification: The OCR recognition of the business license "Legal representative name: Zhang San, ID number: 440106XXXXXXXXXXXX" matches the files in the marketing management system (100% match rate), and is therefore determined to be consistent;

[0095] Standardization process: Normalization formula is used:

[0096]

[0097] Load data processing (e.g., load 120kW on 2024-10-15, x) min =30kW, x max =200kW, calculated to be x norm =0.51, unify the data format to the [0,1] range, and construct a dataset for evaluating electricity services for small and medium-sized enterprises.

[0098] Step 2: Construct a customized evaluation indicator system for SMEs:

[0099] Based on the characteristics of small and medium-sized enterprises, the typical design of China Southern Power Grid, and the "Electric Power User Service Specification" (GB / T38946-2020), four primary indicators and 12 secondary indicators are constructed. The indicator definitions, calculation logic, and examples are as follows:

[0100]

[0101]

[0102]

[0103]

[0104] Step 3: Multi-dimensional evaluation based on deep learning models:

[0105] 3.1 Feature Extraction:

[0106] Electricity load time series characteristics: Input the preprocessed daily load data (720 dimensions) of the past 30 days into the CNN model, output a 128-dimensional feature vector (e.g., peak load ratio 0.65, load growth slope 0.02kW / day), and then combine it with industry production cycle data to calculate dynamic weights (e.g., October load data weight 0.18, February weight 0.08), and perform weighted correction on the feature vector;

[0107] Service order text features: Inputting "3 service orders in the past year containing 'capacity expansion consultation' and 2 service orders containing 'electricity bill inquiry'" into the BERT model, the output is a 512-dimensional vector, and the keyword features "capacity expansion demand: high, electricity bill sensitivity: medium" are extracted;

[0108] Features of the expansion project drawings: Using the Teigha SDK to parse the DWG file, the equipment parameter features were extracted: "Transformer capacity 800kVA, line cross-section 120mm²". 2 The topology feature is "Power access path: T-connected to pole #12 of 10kV Tianhe line".

[0109] 3.2 Multi-dimensional evaluation calculation:

[0110] Power stability index assessment:

[0111] Dynamic weight application: The weighted and corrected 128-dimensional load feature vector is input into the LSTM model. The model learns the correlation between dynamic weights and load fluctuations through the hidden layer (128 units), such as identifying the contribution of "18% increase in high-weight load data in October" to the probability of short-term capacity expansion.

[0112] Output evaluation values: The model outputs dynamic evaluation values: "Load fluctuation coefficient 92% (medium, which is closer to the production cycle than the original 104%), short-term capacity expansion probability 65% ​​(medium), confidence level 97%", which reduces the error rate by 0.3 percentage points compared to the original fixed weight evaluation.

[0113] Cost adaptability evaluation: Input the following into the random forest model: "Connection cost ratio 2.5%, electricity bill payment timeliness rate 91.7%, electricity price package adaptability 80%", and output the evaluation result: "Cost adaptability: high".

[0114] Project compliance indicator assessment: Input the following criteria into the random forest model: "Engineering drawing compliance rate 90%, equipment installation standardization score 8, and meter model matching degree 80%", and output the assessment result: "Project compliance: Medium";

[0115] Service demand potential indicator assessment: Input the following into the random forest model: "Short-term capacity expansion probability 65%, new energy access demand 85%, exploration response demand score 3", and output the assessment result: "Service demand potential: high".

[0116] 3.3 Result Calibration:

[0117] The "SME Assessment Benchmark Library API" of the marketing management system was called to obtain the benchmark values ​​for 10kV enterprises in the same industry (Manufacturing C31): "Load fluctuation coefficient 95%, connection cost 3%", and then calibrated using the deviation correction formula.

[0118] After calibration, the load fluctuation coefficient = 92% × (95% / 100%) = 87.4% (industry average 100%);

[0119] The cost of power connection after calibration accounts for 2.5% × (3% / 2.8%) = 2.7% (industry average 2.8%).

[0120] The post-calibration evaluation error rate is 0.2% ≤ 0.5%, which meets the accuracy requirements. Furthermore, due to the dynamic weight adjustment, the calibration result of the load fluctuation coefficient is closer to the actual production of small and medium-sized enterprises.

[0121] Step 4: Evaluation Results Output and Service Adaptation:

[0122] 4.1 Generate a visual evaluation report:

[0123] Generate a PDF report with the following key content:

[0124] Indicator Score Page: The radar chart shows the scores of four primary indicators: cost adaptability 85 points, power supply stability 78 points (an improvement from the previous 72 points due to more accurate dynamic weighting), engineering compliance 70 points, and service demand potential 88 points, with the note "engineering compliance score below 80 points requires close attention";

[0125] Load weight analysis page: Constructs a line chart to show the dynamic weight distribution of each month in 2024 (October 0.18, November 0.16, December 0.15, February 0.08, March 0.09), with the weight calculation basis (entropy weight method + industry production cycle) and reference to the Southern Power Grid's classic design clauses;

[0126] The engineering rectification page shows a screenshot of a non-compliant item in the engineering drawings: "The transformer grounding resistance should be ≤4Ω, but the actual measurement is 6Ω." It cites Article 4.3.2 of the Southern Power Grid Design Standard and suggests the rectification measure: "Replace the grounding electrode and use two L50×50×2500 galvanized angle irons with a spacing of 5m."

[0127] Service suggestion page: Recommends "Peak-valley time-of-use electricity pricing (peak period 0.75 yuan / kWh, valley period 0.35 yuan / kWh), estimated annual electricity cost savings of 15,000 yuan (due to more accurate load fluctuation assessment, savings of 3,000 yuan more than the original suggestion); recommends connecting a 100kW photovoltaic system to enjoy a subsidy of 0.03 yuan / kWh per kilowatt-hour."

[0128] 4.2 Service Resource Scheduling:

[0129] Exploration personnel matching: The system calls the exploration dispatch business intelligent agent API to filter personnel with the skill tag "high voltage business expansion", distance ≤ 5km, and the lowest saturation (Zhang San, skill: high voltage business expansion, distance 3.2km, saturation 30%). An evaluation report containing "load weight analysis" is pushed to their Elink terminal, and an SMS is sent to the contact person of the small and medium-sized enterprise: "Exploration personnel Zhang San (phone 138XXXX8888) will arrive within 1.5 hours, address: No. XX, XX Road, Tianhe District";

[0130] Power Grid Resource Reservation: Send a POST request to the power grid management platform to reserve 25kVA of transformer capacity in the distribution area based on the "short-term capacity increase probability of 65%" (which is more in line with the dynamic assessment results than the original 20kVA, work order number GY2024005, valid for 7 days). After the platform returns a "reservation successful" receipt, update the work order status to "resource reservation completed".

[0131] 4.3 Service Progress Tracking:

[0132] Progress updates are pushed to SMEs through the Southern Power Grid online platform:

[0133] Real-time synchronization of statuses such as "Survey personnel have departed (current location: XX Road, Tianhe District)" and "Business expansion project rectification completed, awaiting re-inspection";

[0134] After the project rectification review is approved, a notification will be sent stating "Meter installation completed, power expected to be connected the next day". At the same time, you will be informed that "Based on your company's peak production season load characteristics in October, 25kVA capacity has been reserved for you to avoid insufficient power supply during the peak season", ensuring that the service is transparent and meets the actual needs of the company.

[0135] Summarize:

[0136] I. Addressing the issue of assessment dimensions not aligning with the characteristics of SMEs, and improving assessment accuracy:

[0137] Construct a dedicated indicator system (including 12 specific secondary indicators) encompassing cost adaptability, power supply stability, engineering compliance, and service demand potential:

[0138] On the cost side (the proportion of electricity connection costs and the compatibility of electricity price packages), it directly matches the cost-sensitive needs of small and medium-sized enterprises and can push installment payment and other solutions in a targeted manner;

[0139] The load weight is dynamically adjusted by industry production cycle and entropy weight method (e.g., 0.18 for October peak season and 0.08 for February off-season in manufacturing) to make the load fluctuation coefficient assessment more in line with reality.

[0140] On the demand side (probability of short-term capacity expansion and demand for new energy access), the demand for capacity expansion can be predicted to avoid service delays.

[0141] II. Addressing the issues of reliance on manual assessments and low accuracy, thereby improving assessment efficiency and reliability:

[0142] Multi-source data intelligent preprocessing: By connecting to 7 types of systems through RESTful API, combined with missing data filling with the industry average, anomaly removal with 3 times the standard deviation, and consistency verification with OCR + rule engine, manual input errors are eliminated, and high-quality datasets are built.

[0143] Deep learning fusion evaluation: CNN is used to extract workload features, BERT to extract work order text features, and TeighaSDK to extract drawing features. These are then input into an LSTM+random forest model and calibrated against industry benchmarks, resulting in a reduced evaluation error rate and improved confidence.

[0144] It replaces manual assessment (which used to take 15-20 minutes per assessment), significantly improving efficiency.

[0145] III. Addressing the lack of specific system adaptation and difficulties in service implementation in the assessment system, and optimizing service and resource allocation:

[0146] Based on a three-tier, two-link architecture (multi-source integration - intelligent modeling - result application, linking the entire business system and service process):

[0147] Output a PDF visualization report (including indicator radar charts, rectification screenshots and supporting evidence), and simultaneously push out the adaptation plan;

[0148] Survey and dispatch are conducted by skill matching → within 5km → low saturation screening, and power grid resources are reserved according to the probability of capacity expansion to avoid resource waste;

[0149] Service progress is synchronized in real time to the Southern Power Grid online platform and enterprises, improving transparency.

[0150] IV. Achieving a win-win situation for SMEs and power grid companies:

[0151] Small and medium-sized enterprises: reduce electricity costs (avoid inefficient configuration), shorten service response time (such as emergency capacity expansion and speed enhancement), and improve service experience;

[0152] Power grid companies: Optimize resource allocation (more precise load dispatching in distribution areas), reduce operating costs (replace manual assessment), and break the closed loop of inaccurate assessment - inefficient service - resource waste.

Claims

1. A smart assessment method for electricity services for small and medium-sized enterprises, characterized in that, Includes the following steps: Step 1: Multi-source data collection and preprocessing. Through standardized data interfaces, connect with relevant business systems in the power industry, including marketing, power grid management, metering automation, engineering management, business information, service platforms, and human resources, to collect basic attributes, electricity consumption behavior, business expansion projects, service interactions, and exploration resource data of SMEs. After filling missing values, removing outliers, and verifying data consistency, construct an evaluation dataset through standardized processing. Step 2: Construct a dedicated evaluation index system for SMEs. Based on the characteristics of SMEs, such as small electricity consumption, high cost sensitivity, and frequent load fluctuations, and in conjunction with electricity user service standards, establish four primary indicators, including cost adaptability, electricity consumption stability, engineering compliance, and service demand potential, as well as corresponding special secondary indicators that reflect the characteristics of SMEs. Step 3: Multi-dimensional evaluation based on deep learning model. Convolutional neural network, pre-trained language model and engineering drawing analysis technology are used to extract time series, text and engineering features from the evaluation dataset. The features are input into LSTM and random forest fusion model. The load fluctuation dynamic weight mechanism of industry production cycle plus entropy weight method is combined to calculate the initial evaluation value. Then the initial evaluation value is calibrated by calling the evaluation benchmark data of small and medium-sized enterprises in the power industry. Step 4: Output the assessment results and adapt the service. Generate a visual report containing load weight analysis and engineering rectification suggestions according to the power industry assessment report specifications. Based on the assessment results, push the appropriate electricity price package and equipment transformation plan, coordinate the dispatch of survey personnel and reserve grid resources, and synchronize the service progress to the power service platform accessible to small and medium-sized enterprises.

2. The intelligent assessment method for electricity services for SMEs as described in claim 1, characterized in that: In step 1, the standardized data interface is the RESTful API interface; missing value filling is done using the average daily data of small and medium-sized enterprises of the same voltage level in the same industry over the past 3 years; outlier removal is based on the judgment that the data exceeds 3 times the standard deviation of the industry average over the past 3 years; data consistency verification is performed by recognizing business licenses and engineering drawings through Baidu Smart Cloud OCR engine, combined with Drools rule engine verification; standardization processing uses a normalization formula to map the data to the [0,1] interval.

3. The intelligent assessment method for electricity services for SMEs as described in claim 1, characterized in that: The specific secondary indicators in step 2 are as follows: Cost adaptability: percentage of electricity connection cost, on-time electricity payment rate, and compatibility with electricity pricing packages; Power stability category: including load fluctuation coefficient with dynamic weight correction, power outage impact sensitivity, and backup power demand probability; Engineering compliance: compliance rate of engineering drawings, standardization of equipment installation, and matching degree of meter models; Service demand potential category: including short-term capacity expansion probability with dynamic weight adjustment, new energy access demand, and exploration response demand.

4. The intelligent assessment method for electricity services for SMEs as described in claim 1, characterized in that: In step 3, the convolutional neural network has a 3-layer structure, takes the daily load data of the past 30 days as input, and outputs 128-dimensional time-series features; the pre-trained language model is a BERT-base model fine-tuned based on 100,000 power service work orders, and outputs 512-dimensional text features; the engineering drawing parsing technology is TeighaSDK, which extracts equipment parameters through regular expressions and analyzes line topology based on the NetworkX library.

5. The intelligent assessment method for electricity services for SMEs as described in claim 1, characterized in that: The implementation process of the load fluctuation dynamic weighting mechanism in step 3 is as follows: Step 1: Call the industry production cycle query interface to obtain the peak and off-peak seasons of the industries to which SMEs belong, and build a mapping table containing industry codes and cycle periods; The second step is to divide the nearly one year of load data into months, calculate the load entropy value of each month using a formula, and then calculate the basic weight of each month using a formula. Step 3: Adjust the weighting of peak season months by multiplying by 1.2 and off-season months by multiplying by 0.8; Step 4: Use the corrected weights as weighting coefficients for time series features and input them into the LSTM model.

6. The intelligent assessment method for power services for SMEs as described in claim 1, characterized in that: In step 4, the visualization report is in PDF format, including indicator score radar chart, monthly load weight line chart, and screenshots of non-compliant projects; the survey personnel are dispatched according to skill tags → GIS straight-line distance ≤ 5km → work order saturation in ascending order; the power grid resource reservation is based on the short-term capacity expansion probability setting capacity and 7-day validity period; the service progress is synchronized to the Southern Power Grid online platform.

7. A smart power service assessment system for small and medium-sized enterprises, characterized by: The system comprises a multi-source data integration layer, an intelligent assessment and modeling layer, and an assessment result application layer. It also integrates seamlessly with existing power business systems and the entire power service process for SMEs. The multi-source data integration layer connects to marketing management, power grid management, metering automation, engineering management, business information, China Southern Power Grid Online, and human resources systems via RESTful API interfaces to collect and preprocess multi-dimensional data. The intelligent assessment and modeling layer includes a data preprocessing module, a feature engineering module, an LSTM+random forest fusion model integrating a load fluctuation dynamic weight submodule, and a result calibration module. The assessment result application layer includes a report generation module, a service suggestion module, and a resource scheduling module.

8. The intelligent power service assessment system for SMEs as described in claim 7, characterized in that: The collected data includes the skill tags of the surveyors, their WGS84 coordinate system positions, work order saturation, and historical survey completion rates. The report generation module of the application layer synchronizes the report to the Southern Power Grid online platform through the report push API. The resource scheduling module calls the survey dispatch business intelligent agent API to match surveyors and sends a POST request to the power grid management platform to reserve resources. The scheduling results are synchronized to the surveyors' Elink terminals and the contacts of small and medium-sized enterprises.

Images