CN117853269A

CN117853269A - Heat supply data optimization method, device and equipment based on machine learning

Info

Publication number: CN117853269A
Application number: CN202410038551.0A
Authority: CN
Inventors: 王飞飞; 刘大海; 赵永芳; 李晓琴; 刘詟; 孟德凯
Original assignee: Beijing Zhongneng North Technology Co ltd
Current assignee: Beijing Zhongneng North Technology Co ltd
Priority date: 2024-01-10
Filing date: 2024-01-10
Publication date: 2024-04-09
Anticipated expiration: 2044-01-10
Also published as: CN117853269B

Abstract

According to the heat supply data optimization method, the heat supply data optimization device and the heat supply data optimization equipment based on machine learning, due to the fact that the optimized linkage characteristic engineering topology network is provided with the implicit transmission topology pointer, in other words, the heat supply data optimization request task and the heat supply optimization visual strategy to be processed are involved in deeper transmission, the optimized linkage characteristic engineering topology network is complete and accurate in detail output of the multi-source heat supply system state data, the heat supply optimization request task and the heat supply optimization visual strategy to be processed. On the premise that details contained in the linkage characteristic engineering topological network are complete and accurate as much as possible, the heat supply optimization visual strategy to be processed, which is determined from the plurality of heat supply optimization visual strategies to be processed and serves as the optimal heat supply optimization visual strategy, is more reasonable based on the heat supply strategy matching view obtained by each heat supply optimization visual strategy to be processed based on the linkage characteristic engineering topological network.

Description

Heat supply data optimization method, device and equipment based on machine learning

Technical Field

The application relates to the technical field of data processing, in particular to a heat supply data optimization method, device and equipment based on machine learning.

Background

The heating system is a complex energy supply system, and relates to a plurality of energy sources, a transmission network, user demands and other factors. Optimizing the operation of a heating system requires a fine analysis and processing of various status data, heating request tasks and possible heating optimization strategies.

However, conventional optimization methods are typically based on empirical rules or simple statistical models, which are difficult to handle for the complexity and dynamics of these data and tasks. Furthermore, these methods often fail to fully utilize the available information due to the lack of efficient data mining and processing tools, resulting in an undesirable optimization effect.

In recent years, machine learning techniques have demonstrated great potential in many fields. In particular, deep learning, which is capable of extracting useful features and rules from raw data by self-learning and adjustment, thereby significantly improving the ability to deal with complex problems. Therefore, how to apply the machine learning technique to the optimization of the heating system has become an important research direction.

However, despite the remarkable results achieved by machine learning techniques in some fields, challenges remain in how to effectively apply them to optimization of heating systems.

Disclosure of Invention

In order to improve the problems, the application provides a heat supply data optimization method, a heat supply data optimization device and heat supply data optimization equipment based on machine learning.

In a first aspect, a heat supply data optimization method based on machine learning is provided, and is applied to a heat supply data optimization device, and the method includes:

acquiring linkage characteristic engineering topology networks generated by combining multi-source heating system state data, a heating optimization request task and a heating optimization visual strategy to be processed, wherein the linkage characteristic engineering topology networks comprise at least two AI characteristic members and at least one transmission topology pointer connected with the AI characteristic members, the AI characteristic members represent image description information in the multi-source heating system state data, the heating optimization request task and the heating optimization visual strategy to be processed, and the transmission topology pointer represents upstream and downstream transmission characteristics among the image description information;

determining a modeling prediction execution topology corresponding to the linkage characteristic engineering topology network based on the linkage characteristic engineering topology network, wherein the modeling prediction execution topology is used for reflecting a confidence coefficient of an implicit transmission topology pointer between any two AI characteristic members in the linkage characteristic engineering topology network, and the implicit transmission topology pointer represents a transmission topology pointer obtained by mining based on AI characteristic members and front and rear sequence involving vectors of the transmission topology pointer in the linkage characteristic engineering topology network;

Optimizing the linkage characteristic engineering topology network based on the modeling prediction execution topology to obtain an optimized linkage characteristic engineering topology network;

determining a heating strategy matching viewpoint corresponding to the heating optimization visual strategy to be processed according to the state data of the multi-source heating system, the heating optimization request task, the heating optimization visual strategy to be processed and the optimized linkage characteristic engineering topology network, wherein the heating strategy matching viewpoint is used for representing the confidence coefficient of the optimal heating optimization visual strategy corresponding to the heating optimization request task;

and determining the optimal heat supply optimization visual strategy corresponding to the heat supply optimization request task from the at least one heat supply optimization visual strategy to be processed based on the heat supply strategy matching views respectively corresponding to the at least one heat supply optimization visual strategy to be processed corresponding to the heat supply optimization request task.

In some aspects, the determining, based on the linkage feature engineering topology network, a modeling prediction execution topology corresponding to the linkage feature engineering topology network includes:

based on the transmission topology pointers included in the linkage characteristic engineering topology network, outputting a global quantized transmission list corresponding to the linkage characteristic engineering topology network, wherein the global quantized transmission list is used for quantitatively outputting the transmission topology pointers included in the linkage characteristic engineering topology network;

Obtaining an original modeling prediction feature list based on the global quantization transfer list and the trusted factor list;

and carrying out minimum one-time cyclic optimization on the original modeling prediction feature list to obtain at least one optimized modeling prediction feature list, wherein each optimized modeling prediction feature list is used for quantitatively outputting modeling prediction execution topology corresponding to the linkage feature engineering topology network.

In some aspects, the linkage characteristic engineering topology network includes at least one of the following types of transmission topology pointers: a transfer topology pointer in which a surface layer involving element exists, a transfer topology pointer in which a derivative involving element exists, and a transfer topology pointer in which a feature cross exists;

the outputting the global quantized transfer list corresponding to the linkage characteristic engineering topology network based on the transfer topology pointer included in the linkage characteristic engineering topology network comprises the following steps:

outputting a multidimensional feature list corresponding to each kind of transfer topology pointer, wherein each list unit in the multidimensional feature list is used for reflecting whether the transfer topology pointer of each kind exists between two AI feature members, the dimension of the multidimensional feature list is P, P represents the number of AI feature members in the linkage feature engineering topology network, and P is a positive integer;

And outputting a global quantized transfer list corresponding to the linkage characteristic engineering topological network based on multidimensional characteristic lists respectively corresponding to various transfer topological pointers included in the linkage characteristic engineering topological network.

In some aspects, the obtaining an original modeling prediction feature list based on the global quantized transfer list and the trusted factor list includes:

randomly extracting quantized values from quantization constraint quadrants conforming to preset statistical conditions to generate the trusted factor list;

processing the trusted factor list through an interval variable value mapping algorithm to obtain a trusted factor list with the interval variable value mapped;

and weighting the global quantization transfer list and the trusted factor list mapped by the completed interval variable value to obtain the original modeling prediction characteristic list.

In some aspects, the performing a minimum one-cycle optimization on the original modeling prediction feature list to obtain at least one optimized modeling prediction feature list includes:

in the process of the optimization of the ith cycle, weighting an optimized modeling prediction feature list obtained by the optimization of the ith-1 th cycle with the original modeling prediction feature list to obtain an optimized modeling prediction feature list obtained by the optimization of the ith cycle;

And when u=1, the optimized modeling prediction feature list obtained by the u-1 th cycle optimization is the original modeling prediction feature list.

In some aspects, the optimizing the linkage feature engineering topology network based on the modeling prediction execution topology to obtain an optimized linkage feature engineering topology network includes:

for each optimized modeling prediction feature list, configuring a list unit with a list value not smaller than a threshold value in the optimized modeling prediction feature list into a first variable value, and configuring a list unit with a list value smaller than the threshold value into a second variable value to obtain a variable feature list corresponding to the optimized modeling prediction feature list;

and adding the implicit transfer topology pointer between two AI feature members corresponding to the list units with each list value of the variable feature list being the first variable value to obtain the optimized linkage feature engineering topology network.

In some aspects, the determining a heating strategy matching view corresponding to the heating optimization visual strategy to be processed according to the multi-source heating system state data, the heating optimization request task, the heating optimization visual strategy to be processed, and the optimized linkage characteristic engineering topology network includes:

Mining image description vectors corresponding to each image information block in the multi-source heating system state data, the heating optimization request task and the heating optimization visual strategy to be processed through a deep residual error learning model;

determining initial image description vectors respectively corresponding to all AI feature members in the linkage feature engineering topology network based on the image description vectors respectively corresponding to all the image information blocks and the image description information respectively corresponding to all the AI feature members in the linkage feature engineering topology network, wherein each image description information comprises at least one image information block;

determining optimized image description vectors corresponding to all AI feature members in the linkage feature engineering topological network based on the optimized linkage feature engineering topological network and initial image description vectors corresponding to all AI feature members in the linkage feature engineering topological network respectively through a cavity convolution model;

optimizing the image description vectors respectively corresponding to the image information blocks based on the optimized image description vectors respectively corresponding to the AI feature members in the linkage feature engineering topology network through an image description strengthening model to obtain optimized image description vectors respectively corresponding to the image information blocks;

Based on the optimized image description vectors respectively corresponding to the image information blocks, determining the state data of the multi-source heating system and the image description vectors respectively corresponding to the heat supply optimizing visual strategies to be processed;

and determining that the heat supply optimizing visual strategy to be processed is a reliable grading value of the most suitable heat supply optimizing visual strategy corresponding to the heat supply optimizing request task based on the state data of the multi-source heat supply system and the image description vectors respectively corresponding to the heat supply optimizing visual strategy to be processed through a strategy matching processing model, and taking the reliable grading value as a heat supply strategy matching view corresponding to the heat supply optimizing visual strategy to be processed.

In some aspects, the acquiring the linkage characteristic engineering topology network generated by combining the state data of the multi-source heating system, the heating optimization request task and the visual strategy of the heating optimization to be processed includes:

excavating cross-mode feature vectors and image detail units in the state data of the multi-source heating system, the heating optimization request task and the heating optimization visual strategy to be processed;

based on the cross-modal feature vector and feature labels existing in the multi-source heating system state data, the heating optimization request task and the heating optimization visual strategy to be processed, disassembling the multi-source heating system state data, the heating optimization request task and the heating optimization visual strategy to be processed to obtain at least two image description information, and taking the image description information as AI feature members in the linkage feature engineering topology network;

When the same image detail unit exists in the two image description information, generating a transmission topology pointer with feature intersection between AI feature members corresponding to the two image description information respectively;

when two image description information belong to the same visual information cluster and the cross-modal feature vector exists between the two image description information, generating a transmission topology pointer with a surface layer involvement element between AI feature members corresponding to the two image description information respectively;

when two image description information belong to the same visual information cluster and the cross-modal feature vector does not exist between the two image description information, generating a transmission topology pointer with derivative involved elements between AI feature members corresponding to the two image description information respectively;

and obtaining the linkage characteristic engineering topology network based on the determined AI characteristic members and the generated transmission topology pointer.

In a second aspect, there is provided a heating data optimizing apparatus comprising a processor and a memory in communication with each other, the processor being arranged to retrieve a computer program from the memory and to implement the method of the first aspect by running the computer program.

In a third aspect, a computer readable storage medium is provided, on which a computer program is stored, which computer program, when run, implements the method of the first aspect.

In a fourth aspect, there is provided a heating data optimizing apparatus comprising:

the topology generation module is used for acquiring a linkage characteristic engineering topology network generated by combining multi-source heating system state data, a heating optimization request task and a to-be-processed heating optimization visual strategy, wherein the linkage characteristic engineering topology network comprises at least two AI characteristic members and at least one transmission topology pointer connected with the AI characteristic members, the AI characteristic members represent image description information in the multi-source heating system state data, the heating optimization request task and the to-be-processed heating optimization visual strategy, and the transmission topology pointer represents upstream and downstream transmission characteristics among the image description information;

the topology determination module is used for determining modeling prediction execution topology corresponding to the linkage characteristic engineering topology network based on the linkage characteristic engineering topology network, the modeling prediction execution topology is used for reflecting the confidence coefficient of an implicit transmission topology pointer between any two AI characteristic members in the linkage characteristic engineering topology network, and the implicit transmission topology pointer represents the transmission topology pointer obtained by mining based on the AI characteristic members and the front and rear sequence involving vectors of the transmission topology pointer in the linkage characteristic engineering topology network;

The feature optimization module is used for optimizing the linkage feature engineering topology network based on the modeling prediction execution topology to obtain an optimized linkage feature engineering topology network;

the strategy matching module is used for determining a heating strategy matching view corresponding to the heating optimization visual strategy to be processed according to the state data of the multi-source heating system, the heating optimization request task, the heating optimization visual strategy to be processed and the optimized linkage characteristic engineering topology network, wherein the heating strategy matching view is used for representing the confidence coefficient of the optimal heating optimization visual strategy corresponding to the heating optimization request task;

the strategy determining module is used for determining the optimal heat supply optimization visual strategy corresponding to the heat supply optimization request task from the least one heat supply optimization visual strategy to be processed based on the heat supply strategy matching views respectively corresponding to the least one heat supply optimization visual strategy to be processed corresponding to the heat supply optimization request task.

In some aspects, the topology determination module determines a modeling prediction execution topology corresponding to the linkage feature engineering topology network based on the linkage feature engineering topology network, including:

performing least one-time cyclic optimization on the original modeling prediction feature list to obtain at least one optimized modeling prediction feature list, wherein each optimized modeling prediction feature list is used for quantitatively outputting modeling prediction execution topology corresponding to the linkage feature engineering topology network;

the linkage characteristic engineering topology network comprises at least one of the following transmission topology pointers: a transfer topology pointer in which a surface layer involving element exists, a transfer topology pointer in which a derivative involving element exists, and a transfer topology pointer in which a feature cross exists; the outputting the global quantized transfer list corresponding to the linkage characteristic engineering topology network based on the transfer topology pointer included in the linkage characteristic engineering topology network comprises the following steps: outputting a multidimensional feature list corresponding to each kind of transfer topology pointer, wherein each list unit in the multidimensional feature list is used for reflecting whether the transfer topology pointer of each kind exists between two AI feature members, the dimension of the multidimensional feature list is P, P represents the number of AI feature members in the linkage feature engineering topology network, and P is a positive integer; based on multidimensional feature lists respectively corresponding to various transmission topology pointers included in the linkage feature engineering topology network, outputting a global quantized transmission list corresponding to the linkage feature engineering topology network;

The obtaining an original modeling prediction feature list based on the global quantization transfer list and the trusted factor list comprises the following steps: randomly extracting quantized values from quantization constraint quadrants conforming to preset statistical conditions to generate the trusted factor list; processing the trusted factor list through an interval variable value mapping algorithm to obtain a trusted factor list with the interval variable value mapped; weighting the global quantization transfer list and the trusted factor list mapped by the completed interval variable value to obtain the original modeling prediction feature list;

the performing at least one cycle optimization on the original modeling prediction feature list to obtain at least one optimized modeling prediction feature list includes: in the process of the optimization of the ith cycle, weighting an optimized modeling prediction feature list obtained by the optimization of the ith-1 th cycle with the original modeling prediction feature list to obtain an optimized modeling prediction feature list obtained by the optimization of the ith cycle; wherein u is a positive integer, and when u=1, the optimized modeling prediction feature list obtained by the u-1 th cycle optimization is the original modeling prediction feature list;

The feature optimization module optimizes the linkage feature engineering topology network based on the modeling prediction execution topology to obtain an optimized linkage feature engineering topology network, and the feature optimization module comprises the following steps: for each optimized modeling prediction feature list, configuring a list unit with a list value not smaller than a threshold value in the optimized modeling prediction feature list into a first variable value, and configuring a list unit with a list value smaller than the threshold value into a second variable value to obtain a variable feature list corresponding to the optimized modeling prediction feature list; and adding the implicit transfer topology pointer between two AI feature members corresponding to the list units with each list value of the variable feature list being the first variable value to obtain the optimized linkage feature engineering topology network.

In some schemes, the policy matching module determines a heating policy matching view corresponding to the heating optimization visual policy to be processed according to the multi-source heating system state data, the heating optimization request task, the heating optimization visual policy to be processed, and the optimized linkage characteristic engineering topology network, and includes:

In some schemes, the topology generation module obtains a linkage characteristic engineering topology network generated by combining multi-source heating system state data, a heating optimization request task and a to-be-processed heating optimization visual strategy, and the linkage characteristic engineering topology network comprises:

According to the heat supply data optimization method, device and equipment based on machine learning, after the linkage characteristic engineering topology network is generated based on the association situation of the multi-source heat supply system state data, the heat supply optimization request task and the heat supply optimization visual strategy to be processed, further, the confidence coefficient of the implicit transfer topology pointer among all AI characteristic members is determined based on the linkage characteristic engineering topology network. After the confidence coefficient of the implicit transfer topology pointer exists among the AI feature members is obtained, the implicit transfer topology pointer is generated for the AI feature members in the linkage feature engineering topology network, and the optimized linkage feature engineering topology network is obtained. In view of the fact that the optimized linkage characteristic engineering topology network is provided with an implicit transmission topology pointer, in other words, the linkage characteristic engineering topology network comprises deeper transmission involvement conditions among the state data of the multi-source heating system, the heating optimization request task and the heating optimization visual strategy to be processed, the optimized linkage characteristic engineering topology network is more complete and accurate in detail output of the state data of the multi-source heating system, the heating optimization request task and the heating optimization visual strategy to be processed. On the premise that details contained in the linkage characteristic engineering topological network are complete and accurate as much as possible, the heat supply optimization visual strategy to be processed, which is determined from the plurality of heat supply optimization visual strategies to be processed and serves as the optimal heat supply optimization visual strategy, is more reasonable based on the heat supply strategy matching view obtained by each heat supply optimization visual strategy to be processed based on the linkage characteristic engineering topological network.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered limiting the scope, and that other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a flowchart of a heat supply data optimization method based on machine learning according to an embodiment of the present application.

Detailed Description

In order to better understand the technical solutions described above, the following detailed description of the technical solutions of the present application is provided through the accompanying drawings and specific embodiments, and it should be understood that the specific features of the embodiments and embodiments of the present application are detailed descriptions of the technical solutions of the present application, and not limit the technical solutions of the present application, and the technical features of the embodiments and embodiments of the present application may be combined with each other without conflict.

Fig. 1 shows a machine learning based heating data optimization method applied to a heating data optimization apparatus, the method comprising the following steps 110-150.

And 110, acquiring linkage characteristic engineering topology networks generated by combining the state data of the multi-source heating system, the heating optimization request task and the visual strategy of the heating optimization to be processed.

The linked characteristic engineering topology network comprises at least two AI characteristic members and at least one transmission topology pointer connected with the AI characteristic members, wherein the AI characteristic members represent image description information in the multi-source heating system state data, the heating optimization request task and the heating optimization visual strategy to be processed, and the transmission topology pointer represents upstream and downstream conduction characteristics among the image description information.

And 120, determining modeling prediction execution topology corresponding to the linkage characteristic engineering topology network based on the linkage characteristic engineering topology network.

The modeling prediction execution topology is used for reflecting the confidence coefficient that an implicit transmission topology pointer exists between any two AI feature members in the linkage feature engineering topology network, and the implicit transmission topology pointer represents the transmission topology pointer obtained by mining based on the AI feature members and the front and rear sequence involving vectors of the transmission topology pointer in the linkage feature engineering topology network.

And 130, optimizing the linkage characteristic engineering topology network based on the modeling prediction execution topology to obtain an optimized linkage characteristic engineering topology network.

And 140, determining a heating strategy matching view corresponding to the heating optimization visual strategy to be processed according to the state data of the multi-source heating system, the heating optimization request task, the heating optimization visual strategy to be processed and the optimized linkage characteristic engineering topology network.

The heat supply strategy matching view is used for representing that the heat supply optimization visual strategy to be processed is a confidence coefficient of the optimal heat supply optimization visual strategy corresponding to the heat supply optimization request task.

And 150, determining the most suitable heat supply optimization visual strategy corresponding to the heat supply optimization request task from the least one heat supply optimization visual strategy to be processed based on the heat supply strategy matching views respectively corresponding to the least one heat supply optimization visual strategy to be processed corresponding to the heat supply optimization request task.

In step 110, the multi-source heating system status data refers to current operating status information of the multi-source heating system (possibly including solar energy, electric power, gas, etc.), such as temperature, pressure, power, energy consumption, etc. The heat supply optimization request task refers to a requirement for optimizing a heat supply system, such as reducing energy consumption, improving heat supply efficiency, etc., which is set forth by a user or a system manager. The heat supply optimizing visual strategy to be processed is a heat supply strategy to be optimized generated by a system, and the strategy can be displayed in a visual mode, so that the user can conveniently understand and operate the strategy. The linkage characteristic engineering topology network is a network structure which is generated by AI technology according to the state data of the heating system, the heating optimization request task and the visual strategy of the heating optimization to be processed and describes the relation among all factors. The AI feature members are nodes representing a specific attribute or state, such as heat supply temperature, energy consumption, etc., in the linkage feature engineering topology network. The transfer topology pointer is an arrow or line in the linked feature engineering topology network that connects the AI feature members and represents the relationship between them. For example, if an increase in heating temperature may result in an increase in energy consumption, there may be a transfer topology indicator from "heating temperature" to "energy consumption". The image description information contains graphic or image information related to the system state, the policy, such as a chart of the system state, a flowchart of the policy, etc. Upstream and downstream conduction characteristics are used to describe the relationships and directions of influence between AI characteristic members in a linked characteristic engineering topology network, for example, upstream heating temperatures can affect downstream energy consumption.

In some exemplary descriptions, for example, a heating system comprising both solar energy and gas energy is being optimized. The system status data may include solar panel temperature, solar radiation intensity, gas consumption, indoor temperature, etc. Meanwhile, a heat supply optimizing request task is required to reduce the fuel gas consumption to the greatest extent, and the heat supply optimizing visual strategy to be processed may comprise measures such as increasing the solar energy utilization rate, adjusting the heat supply time and the like.

In this case, the linkage feature engineering topology network may be expressed as a network structure of:

(1) AI features some nodes may include "solar panel temperature", "solar radiation intensity", "gas consumption", "indoor temperature", "solar usage", "heating time", etc.;

(2) The transfer topology pointers some arrows or lines may connect different AI feature members. For example, "solar radiation intensity" may affect "solar panel temperature", further affecting "solar utilization"; the solar energy utilization rate and the heating time can jointly influence the fuel gas consumption;

(3) Image description information: the status of each AI feature member may be graphically or otherwise visually represented, such as a time series plot of solar panel temperature, a histogram of gas consumption, etc.;

(4) The upstream and downstream conductive features represent the relationships and directions of influence between the individual AI feature members. In the example, "solar radiation intensity" and "heating time" are upstream features that affect "solar utilization"; whereas "solar energy usage" is an upstream feature of "gas consumption".

The linkage characteristic engineering topology network can clearly draw the operation condition of the system, and provides a basis for further optimization.

In step 120, the modeled predicted execution topology is based on a linked feature engineering topology network, the structure used to make predictions and decisions. For example, in a heating system, this topology may contain elements of heating stations, transmission pipes, users, etc., and describe the relationship and interactions between them. The implicit transfer topology pointer is a relationship index which is not explicitly shown in the linkage characteristic engineering topology network, but is obtained through analysis and learning. For example, although there is no direct edge between temperature and energy consumption, it may be found that there is an implicit relationship between them: as the temperature increases, the energy consumption increases. In machine learning, confidence coefficients are typically used to represent how well a model determines its predicted outcome. For example, if the model predicts that the heat demand will increase on a tomorrow and gives a confidence coefficient of 0.8, this means that 80% confidence in the model will be that the heat demand will increase. In a topological network, the context involvement vector describes the dependency and direction of influence between elements. For example, in a heating system, the "lead" may be a boiler and the "trail" may be a user, meaning that the operational status of the boiler may affect the heating effect of the user.

In some exemplary application scenarios, there is a heating system comprising two sources of energy: solar energy and electricity. The system needs to decide which energy to use according to weather conditions, energy price and other factors.

In this case, first, a linkage feature engineering topology network may be created to describe the operation state of the system. In this network, there may be the following AI feature members: solar radiation intensity, electricity price, solar panel temperature, electricity consumption and solar energy consumption. And, these feature members are connected by a transfer topology pointer, representing the relationship between them.

Then, a corresponding modeling prediction execution topology can be constructed according to the linkage characteristic engineering topology network. This topology may contain the following:

implicit transfer topology pointers: based on historical data and machine learning algorithms, some relationships may be found that were not explicitly defined. For example, it may be found that as the price of electricity increases, the system tends to use more solar energy even if the solar radiation intensity is not high. This relationship may be represented by an implicit transfer topology pointer.

Confidence coefficient: each implicit transfer topology pointer will have a corresponding confidence coefficient that indicates the degree of trust in that relationship. For example, if there is a large amount of data that demonstrates that an increase in the price of electricity results in more solar energy usage, the corresponding confidence coefficient will be high.

By such modeling predicting the execution topology, it is possible to predict how a system may perform under given conditions. For example, if the open day is a cloudy day, but the price of electricity is expected to rise, the model may predict that the system will use more solar energy. In this way, decisions can be made in advance to optimize the operating efficiency of the system.

In step 130, optimizing refers to improving or adjusting the linked feature engineering topology network to increase its performance or efficiency. Optimization may involve changing network structure (e.g., adding, deleting or moving nodes and edges), adjusting model parameters (e.g., learning rates, regularization coefficients, etc.), or modifying feature selection and processing methods. The goal of optimization is typically to maximize or minimize a particular metric, such as reducing prediction errors, improving system efficiency, etc.

The optimized linkage characteristic engineering topology network is a new linkage characteristic engineering topology network obtained after the optimization process. This new network should better reflect the behavior of the system than the network before optimization and be able to make predictions and decisions more accurately.

For example, the heating system includes a solar panel and a gas boiler. The initial linkage feature engineering topology network may only take into account both temperature and energy consumption. However, in the optimization process, it may be found that the solar radiation intensity is also an important factor, so a new node is added to represent this feature. At the same time, it may be found that there are some unnecessary connections in the original network, for example the gas consumption may not directly affect the temperature of the solar panel, so this edge is eliminated. By means of the changes, a new and optimized linkage characteristic engineering topology network is obtained, the behavior of the system is reflected more accurately, and prediction and decision making can be better facilitated.

In step 140 and step 150, the heating strategy matching view is an evaluation index for measuring the matching degree between the heating optimization visual strategy to be processed and the heating optimization request task. The matching degree is usually calculated by a machine learning model according to the state data of the multi-source heating system, the heating optimization request task, the linkage characteristic engineering topology network and the like, and is expressed in the form of a confidence coefficient. For example, if a policy is fully in compliance with a requested task, its heating policy matching perspective may be close to 1; whereas if a policy is widely separated from the requested task, its matching perspective may be near 0.

The optimal heating optimization visual strategy is a heating strategy which best meets the heating optimization request task in all the heating optimization visual strategies to be processed. This strategy is determined by comparing the heating strategy matching perspectives of the individual strategies. For example, assume three visual strategies for optimizing the heat supply to be treated, with matching perspectives of 0.6, 0.8 and 0.7, respectively. Then the strategy with a matching point of view of 0.8 is the most suitable heating optimization visual strategy.

For example, if the heating optimization request task is to reduce energy consumption, possible pending heating optimization visual strategies include: adjusting heating time, improving equipment efficiency, etc. By calculating the heating strategy matching point of each strategy, it is possible to find that the strategy for adjusting the heating time can better meet the request task, so it is the most suitable heating optimization visual strategy.

Next, the above steps 110 to 150 are exemplarily described by way of a complete specific example.

For example, an urban heating system is now being managed, which uses both solar and gas energy.

Firstly, state data (such as solar radiation intensity, air temperature, fuel gas price and the like) of a multi-source heating system are acquired, a heating optimization request task (such as reducing fuel gas consumption to the greatest extent on the premise of guaranteeing to meet user requirements) and a heating optimization visual strategy to be processed (for example, solar energy is preferentially used in daytime and fuel gas is preferentially used at night). Then, a linked feature engineering topology network is generated, which contains AI feature members (such as solar radiation intensity, air temperature, gas price, etc.) and a transmission topology pointer (which represents the relationship between the features).

And secondly, determining modeling prediction execution topology based on the linkage characteristic engineering topology network. For example, it may be found that when the price of gas is high, the system is more prone to use solar energy even if the solar radiation intensity is not very strong. Such implicit relationship may be represented by an implicit pass topology pointer and its corresponding confidence coefficient may be relatively high.

And then executing the topology optimization linkage characteristic engineering topology network based on modeling prediction. Optimization may include adding new AI feature members, deleting unimportant connections, etc., so that the network more accurately reflects the behavior and response of the system.

And then, according to the state data of the multi-source heating system, the heating optimization request task, the heating optimization visual strategies to be processed and the optimized linkage characteristic engineering topology network, determining the heating strategy matching view of each heating optimization visual strategy to be processed. For example, policy a (solar energy is preferred during the day and gas is preferred at night) may have a matching perspective of 0.8, while policy B (heating is provided in a ratio of 50% solar energy to 50% gas, both during the day and at night) may have a matching perspective of 0.6.

And finally, determining the optimal heat supply optimizing visual strategy corresponding to the heat supply optimizing request task based on the heat supply strategy matching view of each heat supply optimizing visual strategy to be processed. In this example, strategy a is selected as the most appropriate heating optimization visual strategy because it has the highest matching perspective.

By the technical scheme, the heating system can be better understood and managed, the efficiency of the heating system is improved, and the requirements of users are met.

By applying the embodiment, after the linkage characteristic engineering topology network is generated based on the association situation of the state data of the multi-source heating system, the heating optimization request task and the heating optimization visual strategy to be processed, further, the confidence coefficient of the implicit transmission topology pointer among all AI characteristic members is determined based on the linkage characteristic engineering topology network. After the confidence coefficient of the implicit transfer topology pointer exists among the AI feature members is obtained, the implicit transfer topology pointer is generated for the AI feature members in the linkage feature engineering topology network, and the optimized linkage feature engineering topology network is obtained. In view of the fact that the optimized linkage characteristic engineering topology network is provided with an implicit transmission topology pointer, in other words, the linkage characteristic engineering topology network comprises deeper transmission involvement conditions among the state data of the multi-source heating system, the heating optimization request task and the heating optimization visual strategy to be processed, the optimized linkage characteristic engineering topology network is more complete and accurate in detail output of the state data of the multi-source heating system, the heating optimization request task and the heating optimization visual strategy to be processed. On the premise that details contained in the linkage characteristic engineering topological network are complete and accurate as much as possible, the heat supply optimization visual strategy to be processed, which is determined from the plurality of heat supply optimization visual strategies to be processed and serves as the optimal heat supply optimization visual strategy, is more reasonable based on the heat supply strategy matching view obtained by each heat supply optimization visual strategy to be processed based on the linkage characteristic engineering topological network.

Further, according to steps 110 to 150, the technical effects achieved can be explained in more detail as follows:

first, in step 110, state data of the multi-source heating system, a heating optimization request task, and a visual strategy of heating optimization to be processed are collected, and a linkage characteristic engineering topology network is constructed based on the information. This network contains a thorough understanding of key factors of the heating system, such as the status of the heating system and its influence, the heating optimization request tasks and the nature of the heating optimization visual strategy to be processed, etc.

Then, in step 120, the linked feature engineering topology network is further analyzed and the confidence coefficient that there is an implicit transfer topology pointer is determined. These confidence coefficients reflect the strength of the association between the various AI feature members, helping to understand and mine hidden, non-intuitive associations in the system.

Next, in step 130, the linkage feature engineering topology network is optimized according to the confidence coefficient, an implicit transmission topology pointer is generated, and the optimized linkage feature engineering topology network is obtained. This optimization process enables the network to describe the behaviour of the heating system more accurately and improves the accuracy of predictions and decisions.

In step 140, a corresponding heating strategy matching view is determined for each heating optimization visual strategy to be processed according to the optimized linkage characteristic engineering topology network. These matching perspectives help understand which policies are more likely to meet the current heating optimization request task.

Finally, in step 150, the most suitable heating optimization visual strategy is selected from the plurality of pending heating optimization visual strategies based on the heating strategy matching perspective. The optimal strategy is the strategy which can best meet the heat supply optimization request task under the current condition, so that the operation of the heat supply system is more efficient and energy-saving.

Through the series of steps, the operation state of the heating system can be understood and analyzed more deeply, and the prediction and decision can be made more accurately, so that the efficiency of the heating system and the satisfaction of users are improved. Meanwhile, the method has strong universality and can be applied to other similar complex system optimization problems.

In general, by constructing a linked feature engineering topology network, complex relationships between state data of a multi-source heating system, heating optimization request tasks and a visual strategy of heating optimization to be processed are captured. Further, a modeling prediction execution topology is determined based on the network and optimized to improve prediction accuracy and decision efficiency. And then, according to the optimized linkage characteristic engineering topology network, determining a heating strategy matching view for each heating optimization visual strategy to be processed, wherein the view represents the matching degree of the strategy and the current optimization request task. And finally, selecting the strategy with the highest matching point as the optimal heating optimizing visual strategy. By the aid of the technical scheme, a decision process of the heating system can be effectively supported, optimal allocation of resources is facilitated, heating efficiency is improved, and user requirements are met.

In some alternative embodiments, the step 120 determines a modeling prediction execution topology corresponding to the linkage feature engineering topology network based on the linkage feature engineering topology network, including steps 121-123.

Step 121, outputting a global quantized transfer list corresponding to the linkage characteristic engineering topology network based on the transfer topology pointer included in the linkage characteristic engineering topology network, where the global quantized transfer list is used to quantitatively output the transfer topology pointer included in the linkage characteristic engineering topology network.

Step 122, obtaining an original modeling prediction feature list based on the global quantization transfer list and the trusted factor list.

And 123, carrying out least one-time cyclic optimization on the original modeling prediction feature list to obtain at least one optimized modeling prediction feature list, wherein each optimized modeling prediction feature list is used for quantitatively outputting modeling prediction execution topology corresponding to the linkage feature engineering topology network.

In the above embodiment, the global quantized transfer list is a list in which transfer topology pointers in the linked feature engineering topology network are represented quantitatively. The quantization process may help to better understand and process the pointers, for example, by converting to a numerical or matrix form.

Quantization output is the process of converting non-digital information (e.g., images, text, etc.) into digital information. In this context, quantization output may involve converting information in the linked feature engineering topology network into a global quantized delivery list.

The list of trusted factors is a list comprising a plurality of trusted factors, each trusted factor representing the trustworthiness of a particular transfer topology pointer. This list may be used to adjust the value of the global quantized delivery list to reflect the importance of the different delivery topology pointers.

The original modeled predicted feature list is a feature list generated based on the global quantized delivery list and the list of trusted factors. This list contains all the features needed to predict the heating system.

Loop optimization is an iterative optimization process aimed at improving the original list of modeled predicted features. In each iteration, the feature list is adjusted according to some optimization criteria (e.g., prediction accuracy, model complexity, etc.).

The optimized modeling prediction feature list is a feature list after cyclic optimization. Compared with the original modeling prediction feature list, the optimized feature list can reflect the actual behavior of the heating system and provide more accurate prediction.

For example, in managing an urban heating system, the system has both solar energy and gas energy sources. The linkage characteristic engineering topology network can be generated according to the system state data, the heat supply optimization request task and the heat supply optimization visual strategy to be processed. The global quantized delivery list is then output based on the delivery topology pointer. Next, the global quantized transfer list and the trusted factor list are combined to obtain an original modeling prediction feature list. And finally, circularly optimizing the original modeling prediction feature list to obtain an optimized modeling prediction feature list.

According to the technical scheme, the complex situation of the multi-source heating system can be quantitatively represented, and a more accurate prediction characteristic list is obtained through optimization, so that the heating system can be better understood and managed, and the efficiency and the user satisfaction of the heating system are improved. Meanwhile, the method has strong universality and can be applied to other similar complex system optimization problems.

In some examples, the linkage feature engineering topology network includes at least one of the following types of transfer topology pointers therein: the existence of the transmission topology pointer of the surface layer involving element, the transmission topology pointer of the derivative involving element and the transmission topology pointer of the characteristic crossing.

Based on this, the global quantized transfer list corresponding to the linkage feature engineering topology network is output in step 121 based on the transfer topology pointer included in the linkage feature engineering topology network, including steps 1211-1212.

Step 1211, outputting a multidimensional feature list corresponding to each kind of transfer topology pointer, where each list unit in the multidimensional feature list is used to reflect whether there is the transfer topology pointer of the kind between two AI feature members, the dimension of the multidimensional feature list is p×p, P represents the number of AI feature members in the linkage feature engineering topology network, and P is a positive integer.

And 1212, outputting a global quantized transfer list corresponding to the linkage characteristic engineering topological network based on multidimensional characteristic lists respectively corresponding to various transfer topological pointers included in the linkage characteristic engineering topological network.

In the above embodiment, the transfer topology pointer in which the surface layer involved elements are present is a transfer topology pointer indicating that there is a clear, direct relationship between AI feature members. For example, there may be such a relationship between gas price and gas consumption.

The existence of a transfer topology pointer that derives the involved elements indicates that an indirect, implicit relationship exists between AI feature members. For example, there may be a relationship between air temperature and solar heating efficiency because although they are not directly related, air temperature may affect solar radiation intensity and thus solar heating efficiency.

The presence of a feature-crossing transfer topology pointer indicates that there is a complex nonlinear relationship between two or more AI feature members. For example, gas prices and air temperatures may have an interactive impact on gas consumption.

A multidimensional feature list is a list whose dimension is equal to the number of AI feature members to indicate whether there are some type of transfer topology pointers between the AI feature members.

The list element is an element in the multidimensional feature list that indicates whether there is some type of transfer topology pointer between two AI feature members.

The size represents the size of the multidimensional feature list, which is equal to the number of AI feature members.

For example, there is a heating system whose linked signature engineering topology network comprises three AI signature members: solar radiation intensity, air temperature and gas price. Thus, the size of the multidimensional feature list corresponding to each type of transfer topology pointer is 3*3. In step 1211, a list of multidimensional features corresponding to each type of delivery topology pointer is output. Then, in step 1212, a global quantized delivery list is output based on the multidimensional feature list corresponding to all types of delivery topology pointers.

By the aid of the technical scheme, the behavior of the heating system can be reflected more accurately, the system state and the optimization decision can be better understood, accordingly, heating efficiency is improved, energy is saved, and user satisfaction is improved.

Under some possible design considerations, the step 122 obtains an original modeling prediction feature list based on the global quantized transfer list and the trusted factor list, including steps 1221-1223.

And 1221, randomly extracting quantized values from the quantized constraint quadrants meeting preset statistical conditions to generate the trusted factor list.

Step 1222, processing the trusted factor list through an interval variable value mapping algorithm to obtain a trusted factor list with the interval variable value mapped.

And step 1223, weighting the global quantization transfer list and the trusted factor list mapped by the completed interval variable value to obtain the original modeling prediction feature list.

In the above embodiment, the preset statistical condition is a set of criteria or requirements preset when performing data analysis or modeling. These conditions may include statistical properties of the distribution, mean, standard deviation, etc. of the data.

A quantization constraint quadrant is a region that defines a particular quantization value range and constraint. For example, in some cases, a quantization constraint quadrant may be defined that contains only positive numbers.

The quantized values are the result of processing and converting non-digital information (e.g., images, text, etc.) into numbers.

The interval variable value mapping algorithm is an algorithm for processing data, which can map an original data value to a specified range or interval.

The trusted factor list for completing interval variable value mapping is the trusted factor list processed by the interval variable value mapping algorithm. Each value in the list has been mapped to a specified range or interval.

Weighting is a process in which the importance or influence of individual elements is different. Weighting is typically achieved by multiplying each element by a corresponding weight.

For example, when managing an urban heating system, first, a list of trusted factors is generated by arbitrarily extracting quantized values from the quantization constraint quadrants that satisfy a preset statistical condition. And then processing the list through an interval variable value mapping algorithm to obtain a trusted factor list for completing interval variable value mapping. And finally, weighting the global quantization transfer list and the trusted factor list with which the interval variable value mapping is completed to obtain an original modeling prediction feature list.

In some examples, in step 1223, the global quantized transfer list is weighted with a list of confidence factors mapped to interval variable values, resulting in an original modeled prediction feature list. Specifically, each raw modeled prediction feature can be calculated by the following formula:

Original modeling prediction feature [ i ] = global quantized transfer list [ i ] list of trusted factors [ i ]

Where i refers to a reference amount or index for locating a specific element in the global quantization delivery list and the list of trusted factors.

For example, if there is a global quantized delivery list [0.8,0.6,0.7] and a trusted factor list [0.5,0.4,0.3] (assuming interval variable value mapping has been completed), then the original modeled prediction feature list can be calculated by the above formula:

original modeling prediction feature [0] =0.8x0.5=0.4;

original modeling prediction feature [1] =0.6x0.4=0.24;

original modeling prediction feature [2] =0.7x0.3=0.21.

Therefore, the original list of modeling prediction features is [0.4,0.24,0.21].

The method can ensure that the importance (represented by a trusted factor) of each transfer topology pointer can be fully considered when an original modeling prediction feature list is generated, so that more accurate and valuable modeling prediction features are obtained.

In this way, the quality of the data is controlled by using preset statistical conditions and quantization constraint quadrants, and then the accuracy and the effectiveness of the data are further improved by an interval variable value mapping algorithm and a weighting operation. The modeling prediction feature list obtained in this way can reflect the actual situation of the heating system more accurately, thereby improving the accuracy of prediction, helping to manage the heating system better, improving the efficiency and meeting the demands of users. Meanwhile, the method has strong universality and can be applied to other similar complex system optimization problems.

In some preferred embodiments, performing at least one loop optimization on the original modeling prediction feature list in step 123 to obtain at least one optimized modeling prediction feature list, including: in the process of the optimization of the ith cycle, weighting an optimized modeling prediction feature list obtained by the optimization of the ith-1 th cycle with the original modeling prediction feature list to obtain an optimized modeling prediction feature list obtained by the optimization of the ith cycle; and when u=1, the optimized modeling prediction feature list obtained by the u-1 th cycle optimization is the original modeling prediction feature list.

In this technical scheme, "loop optimization" refers to an iterative process, in each iteration, the modeling prediction feature list obtained last time after optimization is weighted with the original modeling prediction feature list, so as to obtain a new modeling prediction feature list after optimization. Specifically, assuming that the original modeling prediction feature list is l_0, in the u-th loop optimization, a new optimized modeling prediction feature list l_u is obtained.

The calculation formula can be expressed as: l_u=k l_ (u-1) + (1-k) l_0;

Where k is a weight parameter between 0 and 1 for controlling the specific gravity of the current optimized modeling prediction feature list and the original modeling prediction feature list.

For example, if the original modeled predicted feature list is [0.1,0.2,0.3], at the first round of optimization, the new optimized modeled predicted feature list that may be obtained at the second round of optimization is [0.15,0.25,0.35] assuming k=0.5, assuming the resulting optimized modeled predicted feature list is [0.2,0.3,0.4 ].

In this way, the original modeled predicted feature list is continually optimized, each time the feature list is adjusted based on the previous results. The process can help find a modeling prediction feature list which can better reflect the actual behavior of the heating system, thereby improving the prediction accuracy, helping to better manage the heating system, improving the efficiency and meeting the user demands. Meanwhile, the method has strong universality and can be applied to other similar complex system optimization problems.

In some possible embodiments, the step 130 of optimizing the linkage feature engineering topology network based on the modeling prediction execution topology to obtain an optimized linkage feature engineering topology network includes steps 131-132.

Step 131, for each optimized modeling prediction feature list, configuring a list unit with a list value not smaller than a threshold value in the optimized modeling prediction feature list to be a first variable value, and configuring a list unit with a list value smaller than the threshold value to be a second variable value, so as to obtain a variable feature list corresponding to the optimized modeling prediction feature list.

And 132, adding the implicit transfer topology pointer between two AI feature members corresponding to the list units with each list value of the variable feature list being the first variable value, so as to obtain the optimized linkage feature engineering topology network.

The main objective of steps 131 and 132 is to execute a topology optimization linkage feature engineering topology network based on modeling predictions and generate an optimized linkage feature engineering topology network. Specifically, threshold judgment is carried out on the list value in each optimized modeling prediction feature list, the list value is converted into a variable feature list, and an implicit transfer topology pointer is additionally arranged where needed.

For example, assume there is a list of optimized modeling prediction features: 0.4,0.24,0.21] and sets the threshold to 0.3. Then, list elements having a list value of not less than 0.3 are configured as a first variable value (e.g., set to 1), and list elements having a list value of less than 0.3 are configured as a second variable value (e.g., set to 0). Thus, a new variable feature list is obtained: [1,0,0].

Next, in step 132, an implicit transfer topology pointer is added between two AI feature members corresponding to the list element with each list value being the first variable value in the variable feature list. In this example, only the first list element has a value of 1, so an implicit transfer topology pointer is added between the corresponding two AI feature members.

The method can effectively improve the precision and efficiency of the optimization of the heating system. By thresholding and variating the feature list, those factors that are less important (i.e., less influencing) can be filtered out, focusing more on those factors that are important, having significant influence. Meanwhile, by additionally arranging an implicit transmission topology pointer at a required place, the accuracy and the effectiveness of the linkage characteristic engineering topology network can be further improved, and therefore better support is provided for the optimization decision of the heating system.

Under some preferred design ideas, determining a heating strategy matching view corresponding to the heating optimization visual strategy to be processed in step 140 according to the multi-source heating system state data, the heating optimization request task, the heating optimization visual strategy to be processed and the optimized linkage characteristic engineering topology network, wherein the heating strategy matching view comprises steps 141-146.

And 141, mining image description vectors corresponding to each image information block in the multi-source heating system state data, the heating optimization request task and the heating optimization visual strategy to be processed through a deep residual error learning model.

And 142, determining initial image description vectors respectively corresponding to the AI feature members in the linkage feature engineering topological network based on the image description vectors respectively corresponding to the image information blocks and the image description information respectively corresponding to the AI feature members in the linkage feature engineering topological network, wherein each image description information comprises at least one image information block.

And 143, determining optimized image description vectors corresponding to the AI feature members in the linkage feature engineering topological network respectively based on the optimized linkage feature engineering topological network and the initial image description vectors corresponding to the AI feature members in the linkage feature engineering topological network through a cavity convolution model.

And 144, optimizing the image description vectors respectively corresponding to the image information blocks based on the optimized image description vectors respectively corresponding to the AI feature members in the linkage feature engineering topological network through an image description strengthening model to obtain the optimized image description vectors respectively corresponding to the image information blocks.

And 145, determining the state data of the multi-source heating system and the image description vectors respectively corresponding to the heat supply optimizing visual strategies to be processed based on the optimized image description vectors respectively corresponding to the image information blocks.

Step 146, determining that the to-be-processed heat supply optimization visual strategy is a reliable grading value of the most suitable heat supply optimization visual strategy corresponding to the heat supply optimization request task based on the state data of the multi-source heat supply system and the image description vectors respectively corresponding to the to-be-processed heat supply optimization visual strategy through a strategy matching processing model, and taking the reliable grading value as a heat supply strategy matching view corresponding to the to-be-processed heat supply optimization visual strategy.

In the above embodiment, the deep residual learning model solves the gradient extinction and explosion problems in the deep neural network by introducing "residual connection". Such models are widely used in many computer vision tasks such as image classification, object detection, etc.

The image description vector is a vector obtained by encoding image information, and contains main features and attributes of an image. Typically, such vectors are extracted from the original image by a deep learning model (e.g., convolutional neural network).

An image information block is a local area or segment of an image that contains visual information of that area. In some cases, processing blocks of image information is more efficient or convenient than processing the entire image.

The hole convolution model is a convolution neural network model that captures a wider range of information by introducing "holes" (also called "blobs") in the convolution kernel to increase the receptive field.

The optimized image description vector is an image description vector improved by an optimization method (such as iterative training, parameter tuning and the like). The optimized vector can generally reflect the features and attributes of the image more accurately.

The policy matching process model is a model for determining the best policy. The matching degree or the trusted scoring value of each strategy can be calculated according to the state data of the heating system and the heating optimization visual strategy to be processed.

The credible grading value is a numerical value calculated by the strategy matching processing model and represents the matching degree of a certain heat supply optimization visual strategy and the current heat supply optimization request task. The higher the score value, the more suitable the policy is considered to be for the current task.

For example, there is a set of multi-source heating system status data and a set of heating optimization visual policies to be processed. First, image information blocks are extracted from these data and strategies by a depth residual learning model, and corresponding image description vectors are generated. And then, optimizing the vectors by using a cavity convolution model and an image description enhancement model to obtain an optimized image description vector. Next, an image description vector for each heating optimization visual strategy is determined based on the optimized image description vector. And finally, calculating the credible grading value of each strategy through a strategy matching processing model, and selecting the strategy with the highest grading value as the heating optimization visual strategy which is most suitable for the current task.

The method can effectively select the optimal strategy from a large number of heat supply optimizing visual strategies, thereby helping to better manage a heat supply system, improving heat supply efficiency, saving energy and improving user satisfaction. Meanwhile, the method is based on deep learning and image processing technology, so that the method has strong universality and can be applied to other similar complex system optimization problems.

In other possible embodiments, the step 110 of obtaining the linkage feature engineering topology network generated by combining the multi-source heating system state data, the heating optimization request task and the pending heating optimization visual policy includes steps 111-116.

And step 111, mining cross-modal feature vectors and image detail units in the multi-source heating system state data, the heating optimization request task and the heating optimization visual strategy to be processed.

And 112, disassembling the multi-source heating system state data, the heating optimization request task and the heating optimization visual strategy to be processed based on the cross-mode feature vector and the feature labels existing in the multi-source heating system state data, the heating optimization request task and the heating optimization visual strategy to be processed to obtain at least two image description information, and taking the image description information as AI feature members in the linkage feature engineering topology network.

And 113, when the same image detail unit exists in the two image description information, generating a transmission topology pointer with feature intersection between AI feature members corresponding to the two image description information respectively.

And 114, when two image description information belong to the same visual information cluster and the cross-modal feature vector exists between the two image description information, generating a transmission topology pointer with a surface layer involvement element between AI feature members corresponding to the two image description information respectively.

And 115, when the two image description information belong to the same visual information cluster and the cross-modal feature vector does not exist between the two image description information, generating a transmission topology pointer with derivative involved elements between AI feature members corresponding to the two image description information respectively.

And 116, obtaining the linkage characteristic engineering topology network based on the determined AI characteristic members and the generated transmission topology pointer.

In the above embodiment, the cross-modal feature vector is a feature vector extracted from different types (modalities) of data. For example, if the status data of a heating system includes temperature readings (numeric data) and climate conditions descriptions (text data), then the cross-modal feature vector may include features of both types of data.

In visual information processing, an image detail unit may refer to a specific element or feature in an image, such as color, texture, shape, etc.

Taking an urban heating system as an example, a cross-mode feature vector and an image detail unit are firstly mined from multi-source heating system state data (such as temperature, air pressure, air speed and the like), a heating optimization request task (such as the requirement of a user on indoor temperature) and a heating optimization visual strategy to be processed (such as different energy allocation strategies).

Then, based on the feature vectors and the feature labels, the data and the strategies are disassembled to obtain at least two image description information, and the at least two image description information are used as AI feature members in the linkage feature engineering topology network. For example, if an image detail element of "indoor temperature" is found in both the heating request task and some heating optimization visual strategy, then a transfer topology pointer with feature crossing between their corresponding AI feature members can be generated.

Next, it is decided whether to generate a delivery topology pointer with a surface layer involvement element or a delivery topology pointer with a derivative involvement element between their corresponding AI feature members, depending on whether the two image description information belong to the same visual information cluster and whether a cross-modal feature vector exists between them.

And finally, obtaining the linkage characteristic engineering topology network based on the determined AI characteristic members and the generated transmission topology pointer.

By the method, deep links can be mined from the state data of the multi-source heating system, the heating optimization request task and the visual strategy of the heating optimization to be processed, and an accurate linkage characteristic engineering topology network is constructed by utilizing the links. The method is helpful for better understanding and managing the heating system, improves the accuracy of prediction and decision making, thereby improving the heating efficiency and meeting the demands of users.

In summary, feature analysis is performed on various data generated and acquired in the heating system, key features are identified, the key features are optimized, and the quality, accuracy and usability of the data are improved, so that the data analysis and decision making are better supported.

On the above basis, there is provided a heating data optimizing apparatus including:

The description of the above functional modules may refer to the description of the method shown in fig. 1, and will not be repeated herein.

On the basis of the above, a heating data optimizing apparatus is provided, comprising a processor and a memory in communication with each other, said processor being adapted to retrieve a computer program from said memory and to implement the above-mentioned method by running said computer program.

On the basis of the above, a computer readable storage medium is provided, on which a computer program is stored, which computer program, when run, implements the method described above.

The foregoing is merely exemplary of the present application and is not intended to limit the present application. Various modifications and changes may be made to the present application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. which are within the spirit and principles of the present application are intended to be included within the scope of the claims of the present application.

Claims

1. A machine learning based heating data optimization method, characterized by being applied to a heating data optimization device, the method comprising:

2. The method of claim 1, wherein the determining a modeled predicted execution topology corresponding to the linked feature engineering topology network based on the linked feature engineering topology network comprises:

the performing at least one cycle optimization on the original modeling prediction feature list to obtain at least one optimized modeling prediction feature list includes: in the process of the optimization of the ith cycle, weighting an optimized modeling prediction feature list obtained by the optimization of the ith-1 th cycle with the original modeling prediction feature list to obtain an optimized modeling prediction feature list obtained by the optimization of the ith cycle; and when u=1, the optimized modeling prediction feature list obtained by the u-1 th cycle optimization is the original modeling prediction feature list.

3. The method of claim 2, wherein optimizing the linkage feature engineering topology network based on the modeling prediction execution topology to obtain an optimized linkage feature engineering topology network comprises:

4. The method of claim 1, wherein determining a heating strategy matching perspective corresponding to the heating optimization visual strategy to be processed according to the multi-source heating system state data, the heating optimization request task, the heating optimization visual strategy to be processed, and the optimized linkage feature engineering topology network comprises:

5. The method of claim 1, wherein the obtaining a linkage feature engineering topology network generated in combination with multi-source heating system status data, heating optimization request tasks, and pending heating optimization visual policies comprises:

6. A heating data optimizing apparatus, characterized in that the heating data optimizing apparatus comprises:

7. The heating data optimization device of claim 6, wherein the topology determination module determines a modeled predicted execution topology corresponding to the linked feature engineering topology network based on the linked feature engineering topology network, comprising:

8. The heating data optimizing device according to claim 6, wherein the policy matching module determines a heating policy matching view corresponding to the heating optimizing visual policy to be processed according to the multi-source heating system state data, the heating optimizing request task, the heating optimizing visual policy to be processed, and the optimized linkage characteristic engineering topology network, and the policy matching view comprises:

9. The heating data optimization device of claim 6, wherein the topology generation module obtains a linkage feature engineering topology network generated in combination with multi-source heating system status data, heating optimization request tasks, and a pending heating optimization visual policy, comprising:

10. A heating data optimizing device, characterized in that it comprises a processor and a memory communicating with each other, said processor being arranged to retrieve a computer program from said memory and to implement the method according to any of claims 1-5 by running said computer program.