Data query method and system based on intelligent recommendation type query and early warning

Abstract

The invention provides a data query method and system based on intelligent recommendation type query and early warning, and the method comprises the steps: obtaining an electric power project and a query target inputted by a user, collecting user habit data, working scene data and supplier real-time evaluation data, carrying out the preprocessing of the collected data, and obtaining a data query result; a supplier real-time evaluation splicing feature vector is obtained; the spliced feature vectors are input into a deep learning model, attention weights are dynamically allocated according to the association degree of each feature vector and the query target, and attention features are calculated; carrying out processing operation on the attention features through a hidden layer and an output layer in the deep learning model, and outputting a recommendation result; and based on the recommendation result, calculating a comprehensive early warning index of the supplier, and performing multi-level early warning according to the comprehensive early warning index. According to the method, through intelligent recommendation query and multi-channel early warning based on deep learning, the query efficiency and the accuracy of a query result can be improved, and the risk early warning capability is enhanced.

Description

Technical Field

[0001] This application relates to the field of power data query technology, and in particular to a data query method and system based on intelligent recommendation-based query and early warning. Background Technology

[0002] With the continuous development of the power industry and the advancement of digital transformation, the amount of data related to supplier management in the power industry has grown dramatically. From basic supplier information, such as company size, production capacity, and product types, to performance data and delivery records of the products they provide, as well as service quality feedback during cooperation with power companies, a vast and complex dataset has been formed.

[0003] In related technologies, statistical query methods are typically used to retrieve supplier management data in the power industry. However, these query solutions have several shortcomings given the current massive amounts of data. For example, the query methods are relatively passive, inefficient, and difficult to operate. Furthermore, they lack intelligent recommendation functions and comprehensive early warning mechanisms, which hinders user decision-making and the advancement of power projects. Summary of the Invention

[0004] This application aims to at least partially address one of the technical problems in the related art.

[0005] Therefore, the first objective of this application is to propose a data query method based on intelligent recommendation query and early warning. This method can improve query efficiency and accuracy of query results through intelligent recommendation query based on deep learning and multi-channel early warning, and enhance risk early warning capabilities, which is conducive to optimizing supplier management decisions.

[0006] The second objective of this application is to propose a data query system based on intelligent recommendation-based query and early warning.

[0007] The third objective of this application is to propose an electronic device.

[0008] The fourth objective of this application is to provide a computer-readable storage medium.

[0009] To achieve the above objectives, the first aspect of this application is to propose a data query method based on intelligent recommendation-based query and early warning, comprising the following steps: The system obtains the power project and query target input by the user, and collects various user habit data for supplier management, various work scenario data for the power project, and various real-time evaluation data of suppliers. The collected data is preprocessed to obtain the spliced ​​feature vector of supplier real-time evaluation. The concatenated feature vector is input into a pre-trained deep learning model. Based on the correlation between each feature vector in the concatenated feature vector and the query target, attention weights are dynamically assigned to each feature vector. Attention features are then calculated using the concatenated feature vector and the attention weights. The attention features are processed and calculated using the hidden and output layers of the deep learning model to output recommendation results corresponding to the query target to the user. Based on the recommendation results, a comprehensive early warning index for the supplier is calculated, and multi-level early warnings are issued according to the comprehensive early warning index.

[0010] Optionally, the various user habit data includes query time, query keywords, and operation type and decision type based on the query results. Preprocessing of the various user habit data includes: extracting time features from the query time and converting it into periodic features, wherein the periodic features include query frequency and electricity consumption cycle in the power industry; mapping the query keywords to a low-dimensional vector space using a word vector model to obtain word vectors; performing one-hot encoding on the operation type and the decision type respectively to obtain operation type vectors and decision type vectors; and combining the periodic features, the word vectors, the operation type vectors, and the decision type vectors to generate a user habit feature vector.

[0011] Optionally, the multiple work scenario data includes project type, project size, and project stage. The multiple work scenario data is preprocessed, including: performing one-hot encoding on the project type and the project stage to obtain a project type vector and a stage vector, respectively; standardizing the project size to obtain a size vector; and combining the project type vector, the stage vector, and the size vector to generate a work scenario feature vector.

[0012] Optionally, the various supplier real-time evaluation data includes product quality scores, delivery timeliness scores, after-sales service scores, price competitiveness scores, technical response speed scores, and compliance scores. The various supplier real-time evaluation data are preprocessed, including: normalizing each type of supplier real-time evaluation data to transform each type of supplier real-time evaluation data into a standard range; and combining the various normalized supplier real-time evaluation data to generate a supplier real-time evaluation feature vector.

[0013] Optionally, the deep learning model is a recommendation model based on a multilayer perceptron. The recommendation model includes multiple hidden layers. Each hidden layer is used to calculate the output value corresponding to the input value of the layer using an activation function, a weight matrix formed based on the attention weights, and a bias vector. The output layer includes multiple neurons, each neuron corresponding to a recommendation query. The output layer is used to calculate the probability distribution of each recommendation query using a softmax function, and select the recommendation query with the highest probability as the recommendation result based on the probability distribution.

[0014] Optionally, the calculation of the supplier's comprehensive early warning index includes: constructing an early warning index system based on the product quality score, the delivery timeliness score, the after-sales service score, the price competitiveness score, and multiple financial status indicators of the supplier; assigning a corresponding early warning weight to each indicator in the early warning index system using the analytic hierarchy process (AHP) or principal component analysis (PCA); and adding the products of each indicator in the early warning index system and its corresponding early warning weight to obtain the comprehensive early warning index.

[0015] Optionally, the step of conducting multi-level early warning based on the comprehensive early warning indicator includes: comparing the comprehensive early warning indicator with a preset multi-level early warning threshold, and determining the early warning level corresponding to the comprehensive early warning indicator based on the comparison result; performing time series analysis on the comprehensive early warning indicator to determine the changing trend of the comprehensive early warning indicator, and determining the early warning level based on the changing trend, wherein the early warning level is used to reflect the urgency of the early warning.

[0016] Optionally, the step of conducting multi-level early warning based on the comprehensive early warning indicators further includes: selecting a target early warning channel corresponding to the early warning level from multiple early warning channels, and sending early warning information to the user through the target early warning channel, wherein the multiple early warning channels include the interactive interface of SMS platform, email system and supplier management system.

[0017] To achieve the above objectives, a second aspect of this application also proposes a data query system based on intelligent recommendation-based query and early warning, comprising the following modules: The collection module is used to obtain the power project and query target input by the user, and to collect various user habit data for supplier management, various work scenario data of the power project, and various real-time evaluation data of suppliers. The collected data is preprocessed to obtain the spliced ​​feature vector of supplier real-time evaluation. The calculation module is used to input the concatenated feature vector into a pre-trained deep learning model, dynamically assign attention weights to each feature vector according to the correlation between each feature vector in the concatenated feature vector and the query target, and calculate attention features using the concatenated feature vector and the attention weights. The recommendation module is used to process and compute the attention features through the hidden layers and output layers in the deep learning model, and output recommendation results corresponding to the query target to the user. The early warning module is used to calculate the comprehensive early warning index of the supplier based on the recommendation results, and to issue multi-level early warnings based on the comprehensive early warning index.

[0018] To achieve the above objectives, a third aspect of this application also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a data query method based on intelligent recommendation-based query and early warning as described in any one of the first aspects above.

[0019] To achieve the above objectives, the fourth aspect of this application also proposes a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the data query method based on intelligent recommendation-based query and early warning as described in any one of the first aspects.

[0020] The technical solutions provided by the embodiments of this application bring at least the following beneficial effects: The intelligent recommendation query module based on deep learning can proactively recommend relevant query content to operations personnel based on user habits, work scenarios, and real-time supplier evaluations. This significantly reduces the time and effort operations personnel spend searching for data, improving query efficiency. Simultaneously, the recommendation results more accurately meet the needs of operations personnel in different scenarios, helping them quickly obtain valuable information and make scientific decisions. This improves query efficiency and the accuracy of query results. Furthermore, the comprehensive early warning indicator system and robust early warning triggering mechanism constructed by this application can more accurately identify potential risks from suppliers. By providing detailed early warning information through multiple channels, operations personnel can promptly understand the risk status of suppliers and take timely action based on suggested intervention measures, effectively reducing the impact of supplier risks on power projects and ensuring the smooth progress of power projects. This enhances risk early warning capabilities. Moreover, this application can also optimize supplier management decisions. Based on the intelligently recommended query content and accurate early warning information, it provides operations personnel with richer and more comprehensive decision-making basis. Operations personnel can use this information to better evaluate supplier performance, optimize supplier selection, procurement strategy formulation, and other management decisions, improving the overall level of supplier management in the power industry.

[0021] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0022] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a data query method based on intelligent recommendation-based query and early warning proposed in an embodiment of this application; Figure 2 This is a flowchart illustrating a method for collecting and preprocessing user habit data according to an embodiment of this application; Figure 3 This is a flowchart illustrating a method for collecting and preprocessing work scenario data according to an embodiment of this application; Figure 4 This is a flowchart illustrating a method for collecting and preprocessing real-time supplier evaluation data according to an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an MLP-based intelligent recommendation model proposed in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a data query system based on intelligent recommendation-based query and early warning proposed in an embodiment of this application. Detailed Implementation

[0023] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0024] It should be noted that the statistical queries in the relevant embodiments have the following shortcomings: First, the query method is relatively passive. When power company operators query supplier-related data, they need to clearly understand their query intent and be able to accurately construct query statements. For example, when assessing the reliability of a supplier, operators need to manually input conditions such as the supplier name and a specific time period to filter relevant data such as on-time delivery rate and product failure rate from a large amount of data. This method is not only inefficient, but also difficult to operate for personnel unfamiliar with data structures and query languages.

[0025] Secondly, there is a lack of intelligent recommendation capabilities. Faced with massive amounts of supplier data, operations personnel often struggle to uncover potentially valuable information. When selecting new suppliers, traditional systems cannot proactively recommend suitable suppliers or relevant query dimensions based on the company's past purchasing habits, the current project's work scenario, and real-time evaluations of existing suppliers. Operations personnel can only rely on experience or conduct numerous trial queries, leading to a lengthy and potentially unscientific decision-making process.

[0026] Third, the early warning mechanism is inadequate. Traditional early warning systems are often based on simple threshold settings, such as issuing an alert when a supplier's delivery delays exceed a certain number. However, supplier performance is influenced by a variety of factors, including raw material supply, production equipment status, and unexpected events during transportation. Relying solely on thresholds for a single indicator ignores these complex interrelationships, easily leading to false alarms or missed alerts. This fails to provide operators with comprehensive and accurate early warning information, making it difficult for companies to take timely and effective intervention measures, potentially impacting the normal progress of power projects.

[0027] To address this, this application proposes a data query method and system based on intelligent recommendation-based query and early warning, which can improve data utilization efficiency and management decision-making level, thereby solving the problem that the statistical query and early warning schemes in related embodiments are difficult to meet the growing intelligent and refined needs of power industry supplier management.

[0028] The following description, with reference to the accompanying drawings, illustrates a data query method and system based on intelligent recommendation-based query and early warning, as proposed in an embodiment of this application.

[0029] Figure 1 This is a flowchart illustrating a data query method based on intelligent recommendation-based query and early warning proposed in an embodiment of this application, as shown below. Figure 1 As shown, the method includes the following steps: Step S101: Obtain the power project and query target input by the user, and collect various user habit data for supplier management, various work scenario data for power projects, and various real-time evaluation data of suppliers. Preprocess the collected data to obtain the spliced ​​feature vector of supplier real-time evaluation.

[0030] Specifically, this application employs a deep learning-based intelligent recommendation query strategy to recommend query results corresponding to the user's input query content. The data query scenario targeted by this application is a professional management scenario within the power industry. It outputs query results to users through various methods, such as evaluating supplier reliability and optimizing procurement strategies, thereby achieving the core objective of assisting users in making supplier management decisions. The processing logic for outputting query results (i.e., recommending to users) revolves around "combining power project needs (such as transmission projects and emergency projects) with supplier service capabilities (such as technical response speed and compliance)" to achieve the goal of adapting to user decisions. To this end, this application first collects and preprocesses data based on the power project and query target input by the user. The data preprocessing method of this application is designed specifically for the characteristics of the power industry.

[0031] The data collected and preprocessed in this application includes three aspects: user habit data, work scenario data, and real-time supplier evaluation data. The collection and preprocessing process for each type of data is described in detail below.

[0032] To more clearly illustrate the process of collecting and preprocessing user habit data in this application, a specific method proposed in one embodiment of this application will be described below as an example. Figure 2 This is a flowchart illustrating a method for collecting and preprocessing user habit data according to an embodiment of this application. The various user habit data collected in this embodiment include query time, query keywords, and operation and decision types based on the query results, such as... Figure 2 As shown, the method includes the following steps: Step S201: Extract time features from the query time and convert the query time into periodic features, where the periodic features include query frequency and electricity consumption cycle in the power industry.

[0033] It should be noted that this embodiment can utilize the power company's internal information management system to collect operation logs from staff regarding supplier management. The log content includes the query time. t Search keywords k (e.g., supplier name, product type, etc.), operations on the query results (e.g., clicking to view details, exporting data, etc., by operation type). o (This refers to) and subsequent decisions made based on the query results (such as selecting a supplier, adjusting the purchase quantity, etc., categorized by decision type). d express).

[0034] Specifically, this embodiment extracts time features from the query time and transforms them into periodic features P with monthly or quarterly cycles. t Periodicity characteristic P tThe core components consist of two parts: the first part is the basic cycle dimension, i.e., the query frequency per month / quarter; the second part is the industry cycle dimension, which is determined by P. peak The "peak season / off-peak season" attribute is used as a supplement to the time periodicity feature to adapt to the special business scenarios of the power industry.

[0035] For example, calculate the query frequency for each month or quarter. Simultaneously, this can be done using the monthly or quarterly periodic feature P. t Building upon this foundation, the electricity consumption cycle characteristics of the power industry are added. A distinction is made between "peak electricity consumption seasons (e.g., summer months June-August or winter months December-February)" and "off-peak seasons," denoted as a binary feature P. peak (1 = peak season, 0 = off-peak season).

[0036] Step S202: Map the query keywords to a low-dimensional vector space using a word vector model to obtain word vectors.

[0037] Specifically, for query keywords, a word vector model (such as Word2Vec) is used to map them into a low-dimensional vector space. The training window size is 5, and the number of iterations is 10, thus obtaining the word vectors. During the training process, a dictionary specifically for power industry suppliers was built, containing industry terms such as "ultra-high voltage transmission towers" and "intelligent inspection robots" to ensure the accuracy of word vector mapping.

[0038] Step S203: Perform one-hot encoding on the operation type and decision type respectively to obtain the operation type vector and decision type vector.

[0039] Specifically, for operation type and decision type, one-hot encoding is performed separately to transform them into operation type vectors. (like o =1, then ;like o =2 (and so on) and (like d =1, then ;like d =2, then And so on.

[0040] For example, "operation type" o It refers to "operations by operations personnel on supplier data query results," for example, o =1: Corresponds to "Click to view details" (the most basic interactive operation for query results, and is given priority as the first operation type); o =2: Corresponds to "Export Data" (the second typical operation, usually one of the core requirements after a query). "Decision Type" dIt refers to "supplier management decisions made by operations personnel based on query results," for example, d =1: Corresponds to "Select a supplier" (one of the core decision-making objectives for querying supplier data, and is given priority as the first decision type); d =2: This corresponds to "adjusting the purchase quantity" (the second typical decision, a common management action based on supplier performance).

[0041] Step S204: Combine periodic features, word vectors, operation type vectors, and decision type vectors to generate user habit feature vectors.

[0042] Based on the above steps, the final user habit feature vector constructed is: The total dimensions of the vector are: time feature dimension (1, monthly + quarterly frequency) + word vector dimension (128) + operation type encoding dimension (4) + decision type encoding dimension (5) = 139 dimensions.

[0043] To more clearly illustrate the process of collecting and preprocessing work scenario data in this application, a specific method proposed in one embodiment of this application will be described below as an example. Figure 3 This is a flowchart illustrating a method for collecting and preprocessing work scenario data according to an embodiment of this application. The various work scenario data collected in this embodiment include project type, project scale, and project stage, such as... Figure 3 As shown, the method includes the following steps: Step S301: Perform one-hot encoding on the project type and the stage of the project to obtain the project type vector and the stage vector.

[0044] It should be noted that this embodiment collects relevant data for different work scenarios involving supplier management in power projects. The data comes from the power company's project management system, and the collection period is the entire project lifecycle (i.e., from planning to operation). A single project has approximately 20-30 data fields. These specifically include project types. (Including power generation projects, transmission projects, distribution projects, and energy storage projects, with values ​​ranging from {1: power generation, 2: transmission, 3: distribution, 4: energy storage}), project scale (based on investment amount) Unit: RMB 10,000 or number of equipment (Unit: units / sets for measurement) Project stage (Including the planning, construction, operation, and decommissioning phases, with values ​​ranging from {1: planning, 2: construction, 3: operation, 4: decommissioning}) and the project's urgency. (Including emergency, routine, and postponed, with values ​​ranging from {1: emergency, 2: routine, 3: postponed}).

[0045] Specifically, for each project type, one-hot encoding is used to convert it into a vector. For each stage of the project, one-hot encoding is also performed to obtain the vector. .

[0046] Step S302: Standardize the project size to obtain the size vector.

[0047] Specifically, the project size data is standardized using the following formula: , , in, , These are the mean and standard deviation of investment amounts for all projects in the project management system, respectively. , These represent the mean and standard deviation of the number of devices for all projects in the project management system.

[0048] Step S303: Combine the project type vector, stage vector, and scale vector to generate a work scenario feature vector.

[0049] Based on the above steps, the final combination forms the work scene feature vector. .

[0050] To more clearly illustrate the process of collecting and preprocessing supplier real-time evaluation data in this application, a specific method proposed in one embodiment of this application will be described below as an example. Figure 4 This is a flowchart illustrating a method for collecting and preprocessing real-time supplier evaluation data according to an embodiment of this application. The various types of real-time supplier evaluation data collected in this embodiment include product quality scores, delivery timeliness scores, after-sales service scores, price competitiveness scores, technical response speed scores, and compliance scores, such as... Figure 4 As shown, the method includes the following steps: Step S401: Normalize the real-time evaluation data for each type of supplier to transform the real-time evaluation data for each type of supplier into a standard range.

[0051] It should be noted that this embodiment can obtain various evaluation indicators of suppliers in real time from the supplier evaluation system, with a data update frequency of once every 5 minutes. The supplier evaluation indicator fields are 15-20, including product quality scores. q (Maximum score 100 points, assessed by the quality inspection department based on sampling test results), Timeliness of delivery score d time (The full score is 100 points, and the calculation method is as follows) 0) After-sales service rating s(Maximum score 100, calculated based on after-sales response time and problem resolution rate) Price competitiveness score p comp (Out of 100 points, compared with the industry average price) 00 (If the difference is negative, then take 100 points), technical response speed ttech (Unit: hours, refers to the average time it takes for suppliers to resolve technical issues), compliance score (out of 100 points, assessed based on the validity of qualification certificates and compliance with industry standards), etc.

[0052] Specifically, in this embodiment, the various data mentioned above are normalized to fall within the range of [0,1]. The normalization of each scoring data point can be performed using the following formula: , , , , , , Where the subscript is min and max The parameters are the minimum and maximum values ​​of each scoring indicator. The formula for technical response speed is used for inverse normalization, meaning that the shorter the technical response time, the higher the score.

[0053] In practice, this embodiment can synchronize and evaluate data every 5 minutes, and only renormalize the indicators that have changed (the change threshold is...). (points), the formula is ( (The new value is the indicator value); unchanged indicators retain their normalized values. This reduces the amount of computation.

[0054] Step S402: Combine the real-time supplier evaluation data after various normalization processes to generate a real-time supplier evaluation feature vector.

[0055] Specifically, by combining the normalized real-time evaluation data of each supplier as described above, the final supplier real-time evaluation feature vector is obtained. The vector dimension is 6-dimensional.

[0056] Furthermore, this application concatenates the preprocessed vectors from the above embodiments to obtain the supplier real-time evaluation concatenated feature vector. .

[0057] Step S102: Input the concatenated feature vector into the pre-trained deep learning model, dynamically assign attention weights to each feature vector based on the correlation between each feature vector in the concatenated feature vector and the query target, and calculate attention features using the concatenated feature vector and attention weights.

[0058] Specifically, this application pre-constructs a deep learning model and trains it to obtain an intelligent recommendation model. After inputting the obtained concatenated feature vectors into the pre-trained deep learning model, attention weights are dynamically assigned to the concatenated features using the mutual information between the various feature vectors and the recommendation results.

[0059] As one possible implementation, the attention weights for each feature vector can be calculated using the following formula:

[0060]

[0061]

[0062] in, The mutual information between each feature vector and the recommendation result Y (this parameter is used by the user to measure the relevance) is calculated based on the mutual information between the feature vectors and the query result Y, since the recommendation result Y corresponds to the user's query target. This is the adjustment coefficient.

[0063] Furthermore, the attention features are calculated by multiplying each feature vector in the concatenated feature vector by its corresponding attention weight and then summing them. That is, the final attention features can be calculated using the following formula: .

[0064] Step S103: The attention features are processed and calculated through the hidden layer and output layer in the deep learning model to output the recommendation results corresponding to the query target to the user.

[0065] Specifically, the input layer of the deep learning model in this application receives the concatenated feature vectors. After feature extraction and nonlinear transformation through multiple hidden layers, the output layer finally outputs the recommended query content.

[0066] In one embodiment of this application, the deep learning model is a recommendation model based on a multilayer perceptron. The recommendation model includes multiple hidden layers. Each hidden layer is used to calculate the output value corresponding to the input value of the layer using an activation function, a weight matrix formed based on attention weights, and a bias vector. The output layer includes multiple neurons, each neuron corresponding to a recommendation query. The output layer is used to calculate the probability distribution of each recommendation query using a softmax function, and selects the recommendation query with the highest probability as the recommendation result based on the probability distribution.

[0067] Specifically, such as Figure 5 As shown, the intelligent recommendation model in this embodiment adopts a deep learning model based on a multilayer perceptron (MLP). During the computation of the hidden layer, let the... l The input to the hidden layer is The output is The weight matrix is The bias vector is ,but ,in, For the activation function (for example, the ReLU function can be used): (x)=max(0,x)). Through the calculation of multiple hidden layers, the model can automatically learn the complex relationship between input features and recommended query content.

[0068] During the computation of the output layer, the softmax function is used to calculate the probability distribution of the recommended query content. Assuming the output layer has n neurons, corresponding to n types of recommended queries, the output vector... ,in , This is the input to the i-th neuron in the output layer. The output layer selects the query with the highest probability as the recommendation result.

[0069] The recommendation model in this embodiment can employ the cross-entropy loss function during pre-training. ,in m The number of training samples. For the sample i The probability distribution of the actual recommended query content (one-hot encoding form). This represents the probability distribution predicted by the model. During training, the model's weight parameters can be updated using the stochastic gradient descent (SGD) algorithm. and bias parameters The model is trained iteratively until the loss function converges.

[0070] Step S104: Based on the recommendation results, calculate the comprehensive early warning index of the supplier, and issue multi-level early warnings based on the comprehensive early warning index.

[0071] Specifically, this application also implements multi-channel early warning, using multiple channels to issue tiered warnings to users based on the warning level of the comprehensive early warning indicators. The early warning process is linked to the intelligent recommendation query process in the above embodiments; that is, the recommendation results are integrated into the early warning process and early warning information, and the early warning information also influences the recommendation logic in turn.

[0072] In one embodiment of this application, calculating the supplier's comprehensive early warning index includes: constructing an early warning index system based on product quality scores, delivery timeliness scores, after-sales service scores, price competitiveness scores, and multiple financial status indicators of the supplier; assigning corresponding early warning weights to each index in the early warning index system using the analytic hierarchy process (AHP) or principal component analysis (PCA); and adding the products of each index in the early warning index system with its corresponding early warning weights to obtain the comprehensive early warning index.

[0073] Specifically, this embodiment comprehensively considers multiple dimensions of supplier data to construct a comprehensive early warning indicator system. In addition to the direct evaluation indicators such as product quality, delivery timeliness, after-sales service, and price competitiveness mentioned in the previous embodiments, it also considers the supplier's financial status indicators, such as the debt-to-equity ratio. Current ratio Production capacity utilization rate and market share change rate Data such as...

[0074] Then, each indicator in the early warning indicator system is assigned a corresponding weight. For example, the weight of each indicator can be determined using methods such as the Analytic Hierarchy Process (AHP) or Principal Component Analysis (PCA). Furthermore, the comprehensive early warning indicator can be calculated using the following formula: .

[0075] As an example, the weight coefficients of each indicator can be adjusted based on the user's query content and recommendation results. For instance, if the recommendation results output to the user include a certain indicator, then that indicator may be a key focus for the user, and the weight of that indicator can be appropriately increased.

[0076] Furthermore, based on the calculated comprehensive early warning indicators, it is determined whether an early warning has been triggered, and the corresponding early warning level is determined based on the magnitude of the comprehensive early warning indicators.

[0077] In one embodiment of this application, multi-level early warning is performed based on a comprehensive early warning indicator, including: comparing the comprehensive early warning indicator with a preset multi-level early warning threshold, and determining the early warning level corresponding to the comprehensive early warning indicator based on the comparison result; performing time series analysis on the comprehensive early warning indicator to determine the changing trend of the comprehensive early warning indicator, and determining the early warning level based on the changing trend, wherein the early warning level is used to reflect the urgency of the early warning.

[0078] For example, this embodiment pre-sets different levels of warning thresholds. (in, ).when When a yellow alert is triggered, it indicates that the supplier poses a certain risk and requires attention; when When a red alert is triggered, it indicates a high risk to the supplier, requiring immediate intervention. This embodiment also incorporates time series analysis to observe the warning indicators. I The changing trend. For example, calculating the first derivative of the early warning indicator. and second derivative ,like and Even now I Even if the threshold is not exceeded, an early warning signal can still be issued, and the warning level can be determined based on the magnitude of the derivative value.

[0079] Furthermore, warning information is sent through multiple channels. Once a warning is triggered, detailed warning information is sent to operations personnel through various channels.

[0080] In one embodiment of this application, multi-level early warning based on comprehensive early warning indicators further includes: selecting a target early warning channel corresponding to the early warning level from multiple early warning channels, and sending early warning information to the user through the target early warning channel. The multiple early warning channels include the interactive interfaces of SMS platforms, email systems, and supplier management systems.

[0081] For example, a yellow alert indicates a low level of urgency, and a concise alert SMS message can be sent, including the supplier's name, alert level, and key alert indicators. A red alert indicates a high level of urgency, and a detailed alert report can be sent via email. This report includes the supplier's historical performance data, trend charts of alert indicators, possible risk analysis, and recommended intervention measures. Furthermore, to facilitate timely processing, an alert window can pop up on the power company's supplier management system's user interface, displaying real-time monitoring data, relevant data comparison analysis, and process guidance, allowing operations personnel to understand the situation promptly and make informed decisions.

[0082] In summary, the data query method based on intelligent recommendation-based query and early warning in this application embodiment, with its deep learning-based intelligent recommendation query module, can proactively recommend relevant query content to operations personnel based on user habits, work scenarios, and real-time supplier evaluations. This significantly reduces the time and effort operations personnel spend searching for data, improving query efficiency. Simultaneously, the recommendation results more accurately meet the needs of operations personnel in different scenarios, helping them quickly obtain valuable information and make informed decisions. This improves query efficiency and the accuracy of query results. Furthermore, the comprehensive early warning indicator system and robust early warning triggering mechanism constructed by this method can more accurately identify potential risks from suppliers. By providing detailed early warning information through multiple channels, operations personnel can promptly understand the supplier's risk status and take timely action based on suggested intervention measures, effectively reducing the impact of supplier risks on power projects and ensuring the smooth progress of power projects. This enhances risk early warning capabilities. Moreover, this method can optimize supplier management decisions. Based on intelligently recommended query content and accurate early warning information, it provides operations personnel with richer and more comprehensive decision-making basis. Operations personnel can use this information to better evaluate supplier performance, optimize supplier selection, procurement strategy formulation, and other management decisions, improving the overall level of supplier management in the power industry.

[0083] Based on the above embodiments, in order to more clearly and intuitively describe the specific implementation process of the data query method based on intelligent recommendation query and early warning of this application, a specific embodiment in a practical application will be used as an example for illustration below.

[0084] In this embodiment, suppose Xiao Li, a supplier management and operations staff member at a power company, is currently in charge of the "Summer Emergency Power Transmission Project" and initially wants to query information related to "transmission tower suppliers." The recommendation process of the deep learning model utilizes three major features—user habits, work scenario, and real-time supplier evaluation—in conjunction with the MLP model, specifically including the following steps: Step 1: Collect and preprocess the data.

[0085] The model first extracts Xiao Li's user habit data, current work scenario, and real-time evaluation data of candidate suppliers from the enterprise system, and transforms them into an input feature vector.

[0086] Specifically, the user habit features (vector U, 139 dimensions, simplified key dimensions) include the following data: Time characteristics (Pt): The current time is July 2024 (peak electricity season, Ppeak=1). Xiao Li has queried the "power transmission supplier" twice a month for the past 3 months (monthly frequency=2, quarterly frequency=6). Historical keywords: Previously frequently searched for "delayed delivery of power transmission equipment" and "supplier technical rectification", converted into word vectors using an industry dictionary and Word2Vec (simplified to [0.8, 0.2, ...], 128 dimensions); Historical operations (o): After checking the supplier, we often "exported data" (o=2, unique hotspot code is [0,1,0,0]); Historical decision (d): The purchase quantity was adjusted due to delivery issues (d=2, unique heat code is [0,1,0,0,0]).

[0087] Ultimately, it simplifies to U: [Peak Season 1, Monthly Frequency 2, Word Vector..., Operation [0,1,0,0], Decision [0,1,0,0,0]].

[0088] The following data is included regarding the characteristics of the work scenario (vector S): Current project information: Project type: Power transmission project (p_type=2, unique thermal code [0,1,0,0]); Project size: Investment of 50 million yuan (standardized = 0.7, because the industry average is 40 million yuan and the standard deviation is 10 million yuan). Project phase: Construction phase (p_stage=2, one-hot encoding [0,1,0,0]); Project urgency level: Urgent (p_urgency=1).

[0089] Ultimately simplified to S: [Project type [0,1,0,0], Size 0.7, Stage [0,1,0,0], Urgency 1].

[0090] Using the real-time evaluation features of suppliers (vector V, which is 6-dimensional after normalization), the system screens out 3 mainstream power transmission tower suppliers (A, B, and C).

[0091] Step 2: The model processes the data and outputs the recommendation results.

[0092] First, feature concatenation and attention weight allocation are performed.

[0093] The model first concatenates U, S, and V into a total input vector X (U 139 dimensions + S 8 dimensions + V 6 dimensions = 153 dimensions), and then assigns dynamic weights based on the "relevance between features and the recommendation target". Work Scenario S: The current scenario is an "emergency power transmission project," which is most relevant to "checking power transmission tower suppliers," therefore the weight w_S = 0.4. User habit U: Xiao Li previously focused on "delivery + technical issues", which was the second most relevant, with a weight w_U=0.3; Supplier Evaluation V: Supports specific supplier recommendations, with a weight of w_V=0.3; Finally, the attention feature X_att was calculated as 0.3U + 0.4S + 0.3*V.

[0094] Then, the calculation is performed using an MLP model. The simplified calculation logic is as follows: Hidden layer: X_att's hidden layer is activated by 2 layers of ReLU, and automatically learns the correlation rules such as "urgent projects → need to prioritize delivery / technology" and "users are concerned about delays → focus on matching low-latency suppliers"; Output layer: Calculate the probability of the three candidate query terms using the softmax function. Candidate 1: "Timeliness of delivery and technical response speed of A / B suppliers for emergency power transmission projects in Q3 2024" → Probability 0.72; Candidate 2: "Ranking of all power transmission tower suppliers based on compliance and price competitiveness" → Probability 0.21; Candidate 3: "Historical fault records of power transmission suppliers in Q3 2023" → Probability 0.07.

[0095] Finally, the recommendation results are output.

[0096] The model selects the candidate with the highest probability, number 1, and pushes it to Xiao Li. The interface displays: "Recommended query: Supplier A (delivery timeliness 85 / technical response 2 hours) and Supplier B (delivery timeliness 82 / technical response 2.5 hours) suitable for the emergency power transmission project in Q3 2024 - match your past concerns about delivery issues and the current urgent needs of the project. Click to view detailed data and comparison charts." It should be noted that in this embodiment, because the delivery timeliness of C, after normalization, is only 0.6 (due to high delays), it is excluded by the model through the V feature. Furthermore, because Xiao Li's historical decisions (d=2, adjusting purchase quantity) and actions (o=2, exporting delivery data) show that he pays attention to this type of indicator, and the current project is urgent (S feature) requiring guaranteed delivery efficiency, "delivery + technology" is prioritized. Additionally, because the time feature Pt indicates that the current period is the Q3 peak season, the model automatically associates periodic information to avoid recommending outdated data; therefore, the recommendation results include "Q3 2024," improving the accuracy of the recommendation results.

[0097] To achieve the above embodiments, this application also proposes a data query system based on intelligent recommendation-based query and early warning. Figure 6 This is a schematic diagram of the structure of a data query system based on intelligent recommendation-based query and early warning proposed in an embodiment of this application, as shown below. Figure 6 As shown, the system includes: The collection module 100 is used to obtain the power project and query target input by the user, and to collect various user habit data for supplier management, various work scenario data for power projects, and various real-time evaluation data of suppliers. The collected data is preprocessed to obtain the spliced ​​feature vector of supplier real-time evaluation.

[0098] The calculation module 200 is used to input the concatenated feature vector into the pre-trained deep learning model, dynamically assign attention weights to each feature vector according to the correlation between each feature vector in the concatenated feature vector and the query target, and calculate attention features using the concatenated feature vector and attention weights.

[0099] The recommendation module 300 is used to process and compute attention features through the hidden and output layers of the deep learning model, and output recommendation results corresponding to the query target to the user.

[0100] The early warning module 400 is used to calculate the comprehensive early warning index of the supplier based on the recommendation results, and to issue multi-level early warnings based on the comprehensive early warning index.

[0101] It should be noted that the foregoing explanation of the data query method based on intelligent recommendation and early warning also applies to the system of this embodiment, and will not be repeated here.

[0102] In summary, the data query system based on intelligent recommendation-based query and early warning in this application embodiment can improve query efficiency and accuracy of query results, and enhance risk warning capabilities through intelligent recommendation query based on deep learning and multi-channel early warning, which is conducive to optimizing supplier management decisions.

[0103] To implement the above embodiments, this application also proposes an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the data query method based on intelligent recommendation query and early warning as described in any of the first aspect embodiments above.

[0104] To implement the above embodiments, this application also proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the data query method based on intelligent recommendation-based query and early warning as described in any one of the first aspect embodiments above.

[0105] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0106] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0107] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0108] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0109] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0110] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0111] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0112] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A data query method based on intelligent recommendation-based query and early warning, characterized in that, Includes the following steps: The system obtains the power project and query target input by the user, and collects various user habit data for supplier management, various work scenario data for the power project, and various real-time evaluation data of suppliers. The collected data is preprocessed to obtain the spliced ​​feature vector of supplier real-time evaluation. The concatenated feature vector is input into a pre-trained deep learning model. Based on the correlation between each feature vector in the concatenated feature vector and the query target, attention weights are dynamically assigned to each feature vector. Attention features are then calculated using the concatenated feature vector and the attention weights. The attention features are processed and calculated using the hidden and output layers of the deep learning model to output recommendation results corresponding to the query target to the user. Based on the recommendation results, a comprehensive early warning index for the supplier is calculated, and multi-level early warnings are issued according to the comprehensive early warning index.

2. The method according to claim 1, characterized in that, The various user habit data includes query time, query keywords, and operation and decision types based on query results. Preprocessing of this various user habit data includes: The query time is subjected to time feature extraction, and the query time is converted into periodic features, wherein the periodic features include query frequency and electricity consumption cycle in the power industry; The query keywords are mapped to a low-dimensional vector space using a word vector model to obtain word vectors. One-hot encoding is performed on the operation type and the decision type respectively to obtain the operation type vector and the decision type vector; By combining the periodic features, the word vectors, the operation type vectors, and the decision type vectors, a user habit feature vector is generated.

3. The method according to claim 1, characterized in that, The various work scenario data includes project type, project scale, and project stage. Preprocessing of this data includes: One-hot encoding is performed on the project type and the stage of the project to obtain the project type vector and the stage vector, respectively. The project size is standardized to obtain a size vector; By combining the project type vector, the stage vector, and the scale vector, a work scenario feature vector is generated.

4. The method according to claim 1, characterized in that, The various real-time supplier evaluation data include product quality scores, delivery timeliness scores, after-sales service scores, price competitiveness scores, technical response speed scores, and compliance scores. Preprocessing of this data includes: The real-time evaluation data for each type of supplier is normalized to transform the real-time evaluation data for each type of supplier into a standard range. By combining various normalized supplier real-time evaluation data, a supplier real-time evaluation feature vector is generated.

5. The method according to claim 1, characterized in that, The deep learning model is a recommendation model based on a multilayer perceptron. The recommendation model includes multiple hidden layers. Each hidden layer is used to calculate the output value corresponding to the input value of the layer using an activation function, a weight matrix formed based on the attention weights, and a bias vector. The output layer includes multiple neurons, each corresponding to a recommended query. The output layer is used to calculate the probability distribution of each recommended query using a softmax function, and selects the recommended query with the highest probability as the recommendation result based on the probability distribution.

6. The method according to claim 4, characterized in that, The comprehensive early warning indicators for the calculated supplier include: An early warning indicator system is constructed based on the product quality score, the delivery timeliness score, the after-sales service score, the price competitiveness score, and multiple financial indicators of the supplier. The corresponding early warning weights are assigned to each indicator in the early warning indicator system using either the Analytic Hierarchy Process (AHP) or the Principal Component Analysis (PCA). The comprehensive early warning index is obtained by adding the products of each index in the early warning index system and its corresponding early warning weight.

7. The method according to claim 1, characterized in that, The multi-level early warning based on the comprehensive early warning indicators includes: The comprehensive early warning index is compared with a preset multi-level early warning threshold, and the early warning level corresponding to the comprehensive early warning index is determined based on the comparison result. A time series analysis is performed on the comprehensive early warning indicators to determine their changing trends, and an early warning level is determined based on these trends. The early warning level reflects the urgency of the early warning.

8. The method according to claim 7, characterized in that, The method of providing multi-level early warnings based on the comprehensive early warning indicators also includes: Select a target early warning channel corresponding to the early warning level from a variety of early warning channels, and send early warning information to the user through the target early warning channel. The various early warning channels include the interactive interfaces of SMS platform, email system and supplier management system.

9. A data query system based on intelligent recommendation-based query and early warning, characterized in that, Includes the following modules: The collection module is used to obtain the power project and query target input by the user, and to collect various user habit data for supplier management, various work scenario data of the power project, and various real-time evaluation data of suppliers. The collected data is preprocessed to obtain the spliced ​​feature vector of supplier real-time evaluation. The calculation module is used to input the concatenated feature vector into a pre-trained deep learning model, dynamically assign attention weights to each feature vector according to the correlation between each feature vector in the concatenated feature vector and the query target, and calculate attention features using the concatenated feature vector and the attention weights. The recommendation module is used to process and compute the attention features through the hidden layers and output layers in the deep learning model, and output recommendation results corresponding to the query target to the user. The early warning module is used to calculate the comprehensive early warning index of the supplier based on the recommendation results, and to issue multi-level early warnings based on the comprehensive early warning index.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the data query method based on intelligent recommendation query and early warning as described in any one of claims 1-8.

Images