CN117419427B

CN117419427B - Constant temperature and humidity air cabinet control method and system based on intelligent workshop

Info

Publication number: CN117419427B
Application number: CN202311746267.0A
Authority: CN
Inventors: 谢建和
Original assignee: Guangdong Blue Sea Purification And Energy Conservation Technology Co ltd
Current assignee: Guangdong Blue Sea Purification And Energy Conservation Technology Co ltd
Priority date: 2023-12-19
Filing date: 2023-12-19
Publication date: 2024-03-12
Anticipated expiration: 2043-12-19
Also published as: CN117419427A

Abstract

According to the constant temperature and humidity air cabinet control method and system based on the intelligent workshop, the plurality of target matters executed in the multi-target reasoning network are target matters meeting the target matter involvement conditions, namely the multi-target reasoning network in the intelligent workshop determines a composition framework under the condition that the target matters are explicit, so that the multi-target reasoning network obtained by determining the composition framework is more reasonable.

Description

Constant temperature and humidity air cabinet control method and system based on intelligent workshop

Technical Field

The disclosure relates to the field of data processing, and more particularly, to a constant temperature and humidity air cabinet control method and system based on an intelligent workshop.

Background

With the digital transformation and intelligent development of industrial production, the concept of intelligent workshops has been developed. The intelligent workshop connects the sensor, the equipment and the system through introducing technologies such as the Internet of things, cloud computing, big data analysis and the like, and real-time monitoring, remote control and intelligent optimization of the production process are realized. In this context, the constant temperature and humidity cabinet control system has also been upgraded and improved. In an intelligent workshop, a constant temperature and constant humidity wind cabinet control system acquires temperature and humidity data inside and outside a wind cabinet and the running state of equipment in the wind cabinet in real time through a sensor, and transmits the data to a central control system or a cloud platform for processing and analysis. Based on data analysis and modeling, the system can perform target matters such as temperature and humidity change trend reasoning prediction, load change reasoning prediction, equipment fault reasoning prediction, energy consumption reasoning prediction and the like so as to optimize a control strategy based on a predicted result. In order to deal with numerous target matters simultaneously, a multi-target reasoning algorithm (typically a neural network algorithm) is generally adopted in the system to execute the target matters, and for the multiple target matters, determining a reasonable network architecture is a precondition for ensuring data reasoning accuracy.

Disclosure of Invention

In view of this, the embodiments of the present application at least provide a constant temperature and humidity air cabinet control method and system based on an intelligent workshop, so as to ensure accuracy of data reasoning.

According to an aspect of the disclosed embodiments, there is provided a constant temperature and humidity air cabinet control method based on an intelligent workshop, which is applied to a constant temperature and humidity air cabinet control system, the method including: acquiring wind cabinet environment monitoring data corresponding to a target constant temperature and humidity wind cabinet, wherein the wind cabinet environment monitoring data are used for carrying out data reasoning in first target reasoning matters, the wind cabinet environment monitoring data are input into a multi-target reasoning network, the multi-target reasoning network is used for executing a plurality of target matters comprising the first target reasoning matters, the plurality of target matters meet target matter involvement conditions, the multi-target reasoning network comprises a main multi-layer perceptron, a first multi-layer perceptron corresponding to the first target reasoning matters and a first weight distribution module, and the main multi-layer perceptron is used for inputting the output characterization vectors into the plurality of target matters; extracting characterization vectors of the wind cabinet environment monitoring data according to the first multi-layer perceptron and the main multi-layer perceptron to obtain a first characterization vector output by the first multi-layer perceptron and a general characterization vector output by the main multi-layer perceptron; performing characterization vector correction on the general characterization vector according to the first weight distribution module, and integrating the corrected characterization vector with the first characterization vector to obtain a first integrated characterization vector; performing data reasoning on the first integrated characterization vector according to a first reasoning module corresponding to the first target reasoning item to obtain a first reasoning result corresponding to the first target reasoning item; and controlling the target constant temperature and humidity air cabinet according to the first reasoning result.

According to an example of the embodiment of the present disclosure, the plurality of target matters include a second target reasoning matter, the wind cabinet environment monitoring data is further used for data reasoning in the second target reasoning matter, and the multi-target reasoning network includes a second multi-layer sensor and a second weight distribution module corresponding to the second target reasoning matter; the method further comprises the steps of: extracting the characterization vector of the wind cabinet environment monitoring data according to the second multi-layer perceptron to obtain a second characterization vector output by the second multi-layer perceptron; performing characterization vector correction on the general characterization vector according to the second weight distribution module, and integrating the corrected characterization vector with the second characterization vector to obtain a second integrated characterization vector; and carrying out data reasoning on the second integrated characterization vector according to a second reasoning module corresponding to the second target reasoning item to obtain a second reasoning result corresponding to the second target reasoning item.

According to one example of an embodiment of the present disclosure, wherein the plurality of target items includes a temperature target inference item and a humidity target inference item; the data reasoning is performed on the first integrated characterization vector according to a first reasoning module corresponding to the first target reasoning item to obtain a first reasoning result corresponding to the first target reasoning item, including: performing data reasoning on the first integrated characterization vector according to a first reasoning module corresponding to the temperature target reasoning item to obtain a reasoning temperature corresponding to the temperature target reasoning item; the data reasoning is performed on the second integrated characterization vector according to a second reasoning module corresponding to the second target reasoning item to obtain a second reasoning result corresponding to the second target reasoning item, including: and carrying out data reasoning on the second integrated characterization vector according to a second reasoning module corresponding to the humidity target reasoning item to obtain the reasoning humidity corresponding to the temperature target reasoning item.

According to one example of an embodiment of the present disclosure, the network architecture of the multi-objective inference network is determined using the following steps: acquiring first wind cabinet environment monitoring example data corresponding to first target reasoning matters and second wind cabinet environment monitoring example data corresponding to second target reasoning matters, wherein the first wind cabinet environment monitoring example data carries first target matter indication information, the second wind cabinet environment monitoring example data carries second target matter indication information, and the target matter indication information is used for indicating a comparison result of the wind cabinet environment monitoring example data in the target matters; performing characterization vector extraction on the first wind cabinet environment monitoring example data according to a first target item network corresponding to a first target reasoning item to obtain a first characterization vector, and performing characterization vector extraction on the second wind cabinet environment monitoring example data according to a second target item network corresponding to a second target reasoning item to obtain a second characterization vector; determining a implication score between the first target inference item and the second target inference item based on the implication between the first token vector and the second token vector, and the implication between the first target item indication information and the second target item indication information; determining a composition architecture of an alternative multi-objective inference network corresponding to the first objective inference item and the second objective inference item when the implication score meets a target item implication condition; the composition architecture of the alternative multi-target reasoning network comprises an alternative backbone multi-layer sensor, an alternative first multi-layer sensor and an alternative first weight distribution module corresponding to the first target reasoning item, and an alternative second multi-layer sensor and an alternative second weight distribution module corresponding to the second target reasoning item, wherein the alternative backbone multi-layer sensor is used for inputting the output characterization vector into the first target reasoning item and the second target reasoning item.

According to one example of an embodiment of the present disclosure, the determining the implication score between the first target inference item and the second target inference item according to the implication between the first token vector and the second token vector and the implication between the first target item indication information and the second target item indication information includes: respectively carrying out low-dimensional mapping processing on the first characterization vector and the second characterization vector, and determining a first mapping characterization vector corresponding to the first characterization vector and a second mapping characterization vector corresponding to the second characterization vector; determining the implication score according to the implication between the first and second target item indication information when the first and second mapping token vectors meet a commonality metric requirement with each other.

According to an example of the embodiment of the present disclosure, the extracting the characterization vector from the first target item network corresponding to the first target reasoning item to obtain a first characterization vector, and the extracting the characterization vector from the second target item network corresponding to the second target reasoning item to obtain a second characterization vector from the second wind cabinet environment monitoring example data, includes: extracting a characterization vector of the first wind cabinet environment monitoring example data according to a first target item network corresponding to the first target reasoning item to obtain a plurality of first sub-characterization vectors; carrying out characterization vector combination on the plurality of first sub-characterization vectors to obtain a first characterization vector; extracting a characterization vector of the second wind cabinet environment monitoring example data according to a second target item network corresponding to the second target reasoning item to obtain a plurality of second sub-characterization vectors; and carrying out characterization vector combination on the plurality of second sub-characterization vectors to obtain a second characterization vector.

According to an example of the embodiment of the present disclosure, the extracting the characterization vector from the first target item network corresponding to the first target reasoning item to obtain a first characterization vector, and the extracting the characterization vector from the second target item network corresponding to the second target reasoning item to obtain a second characterization vector from the second wind cabinet environment monitoring example data, includes: extracting the characterization vector of the first wind cabinet environment monitoring example data according to a first target item network corresponding to the first target reasoning item to obtain a plurality of first alternative characterization vectors; performing key characterization vector determination on the plurality of first alternative characterization vectors to obtain the first characterization vectors; extracting the characterization vector of the second wind cabinet environment monitoring example data according to a second target item network corresponding to the second target reasoning item to obtain a plurality of second alternative characterization vectors; and determining key characterization vectors of the second alternative characterization vectors to obtain the second characterization vectors.

According to one example of the embodiment of the present disclosure, the first wind cabinet environment monitoring example data includes P wind cabinet environment monitoring example sub-data, and the plurality of first alternative characterization vectors includes Q first alternative characterization vectors, where P is greater than or equal to 1, and Q is greater than 1; performing key token vector determination on the plurality of first alternative token vectors to obtain the first token vectors, including: reasoning the Q first alternative characterization vectors in the first target reasoning matters to obtain reference target reasoning matters results corresponding to the P wind cabinet environment monitoring example sub-data respectively; determining the area under the receiver operation characteristic curve according to the reference target reasoning item results respectively corresponding to the P wind cabinet environment monitoring example sub-data and the target item indication information respectively corresponding to the P wind cabinet environment monitoring example sub-data; reasoning the characterization vectors except for the T-th alternative characterization vector in the Q first alternative characterization vectors in the first target reasoning matters to obtain T-th target reasoning matters corresponding to the P wind cabinet environment monitoring example sub-data respectively; determining the area under an S-th receiver operation characteristic curve according to the reference target reasoning item results respectively corresponding to the P wind cabinet environment monitoring example sub-data and target item indication information respectively corresponding to the P wind cabinet environment monitoring example sub-data, wherein T=Q-S, and S is less than or equal to Q; and determining the key characterization vectors of the Q first alternative characterization vectors according to the difference value between the area under the S-th receiver operation characteristic curve and the area under the reference receiver operation characteristic curve, so as to obtain the first characterization vectors respectively corresponding to the P wind cabinet environment monitoring example sub-data.

According to an example of the embodiment of the present disclosure, a first target item network corresponding to the first target reasoning item includes a characterization vector eccentric adjustment module, where the characterization vector eccentric adjustment module is configured to determine an eccentric coefficient of an extracted characterization vector in a reasoning process of the network; performing key token vector determination on the plurality of first alternative token vectors to obtain the first token vectors, including: inputting the plurality of alternative characterization vectors into the characterization vector eccentric adjustment module, and outputting eccentric coefficients respectively corresponding to the plurality of alternative characterization vectors; and determining key characterization vectors of the plurality of candidate characterization vectors according to the eccentric coefficients respectively corresponding to the plurality of candidate characterization vectors to obtain the first characterization vector.

According to another aspect of the embodiments of the present disclosure, there is provided a constant temperature and humidity wind cabinet control system, including: a processor; and a memory, wherein the memory has stored therein a computer program which, when executed by the processor, causes the processor to perform the method described above.

The application has at least the beneficial effects that: according to the constant temperature and constant humidity air cabinet control method and system based on the intelligent workshop, characterization vector extraction is conducted on air cabinet environment monitoring data according to the first multi-layer perceptrons and the main multi-layer perceptrons in the multi-objective reasoning network, a first characterization vector and a general characterization vector are obtained, the general characterization vector is corrected according to the first weight distribution module, the corrected characterization vector is integrated with the first characterization vector to obtain a first integrated characterization vector, and finally data reasoning is conducted on the first integrated characterization vector according to the first reasoning module to obtain a reasoning result corresponding to a first objective reasoning item. Firstly, a plurality of target items executed in a multi-target reasoning network are target items conforming to the target item involvement conditions, namely, the multi-target reasoning network in the application determines a composition framework under the condition that the target item involvement is clear, so that the multi-target reasoning network obtained by determining the composition framework is more reasonable.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the aspects of the present application.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a schematic architecture diagram of an application scenario of the present application;

FIGS. 2-3 are schematic diagrams illustrating a determining process of a network architecture provided in the present application;

FIG. 4 is a schematic flow chart of a constant temperature and humidity air cabinet control method based on an intelligent workshop;

fig. 5 is a schematic structural diagram of a constant temperature and humidity air cabinet control device according to an embodiment of the present application;

fig. 6 is a schematic structural diagram of a computer device of a constant temperature and humidity air cabinet control system according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, are intended to be within the scope of the present application.

For a clearer understanding of the present application, a schematic diagram of an application scenario implementing the present application is first introduced, and as shown in fig. 1, the application scenario includes a constant temperature and humidity air cabinet control system 10 and a data collection device cluster, where the data collection device cluster may include one or more data collection devices, and the number of the data collection devices will not be limited here. As shown in fig. 1, the data acquisition device cluster may specifically include a data acquisition device 1, a data acquisition device 2, …, and a data acquisition device n; it will be appreciated that the data acquisition device 1, the data acquisition device 2, the data acquisition devices 3 and … and the data acquisition device n can be connected with the constant temperature and humidity air cabinet control system 10 through a network, so that each data acquisition device can perform data interaction with the constant temperature and humidity air cabinet control system 10 through the network connection.

It is understood that the constant temperature and humidity air cabinet control system 10 may refer to a device that performs target item reasoning, and may be a computer device such as a server or a personal PC that can perform data processing. The constant temperature and humidity cabinet control system 10 may also be used to store cabinet environmental monitoring data. When the constant temperature and humidity wind cabinet control system 10 is used as a server, the constant temperature and humidity wind cabinet control system may be an independent physical server, may be a server cluster or a distributed system formed by at least two physical servers, and may also be a cloud server for providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, a content distribution network (Content Delivery Network, CDN), basic cloud computing services such as big data and an artificial intelligent platform. The data acquisition device may specifically be various sensors for detecting the temperature and humidity of the inside and outside of the wind cabinet, or executing mechanisms such as a heater, a refrigeration compressor, a fan, a humidifier, a dehumidifier, etc., but is not limited thereto. The data collection devices and the servers can be directly or indirectly connected through wired or wireless communication modes, and meanwhile, the number of the data collection devices and the servers can be one or at least two, and the application is not limited herein.

According to the constant temperature and constant humidity air cabinet control method and system based on the intelligent workshop, characterization vector extraction is conducted on air cabinet environment monitoring data according to the first multi-layer perceptrons and the main multi-layer perceptrons in the multi-objective reasoning network, a first characterization vector and a general characterization vector are obtained, the general characterization vector is corrected according to the first weight distribution module, the corrected characterization vector is integrated with the first characterization vector to obtain a first integrated characterization vector, and finally data reasoning is conducted on the first integrated characterization vector according to the first reasoning module to obtain a reasoning result corresponding to a first objective reasoning item. Firstly, a plurality of target items executed in a multi-target reasoning network are target items conforming to the target item involvement conditions, namely, the multi-target reasoning network in the application determines a composition framework under the condition that the target item involvement is clear, so that the multi-target reasoning network obtained by determining the composition framework is more reasonable.

It can be understood that, the reasonable network composition architecture is a precondition for ensuring accuracy of data analysis and further accurately completing reasoning of target items, and the embodiment of the present application is described first from a determination process of the network composition architecture, referring to fig. 2, which specifically may include the following steps: step S110, first wind cabinet environment monitoring example data corresponding to a first target reasoning item and second wind cabinet environment monitoring example data corresponding to a second target reasoning item are obtained.

The first wind cabinet environment monitoring example data carries first target item indication information, the second wind cabinet environment monitoring example data carries second target item indication information, and the target item indication information is used for indicating a comparison result of the wind cabinet environment monitoring example data in target items, and the comparison result can be characterized as a label or a mark.

In one embodiment, the number of the first wind cabinet environment monitoring example data may be one or more, and the number of the wind cabinet environment monitoring example data may be one or more. The environmental monitoring example data of the wind cabinet records relevant data of the constant temperature and humidity wind cabinet, such as various sensor data for detecting the temperature and humidity of the inside and outside of the wind cabinet, and equipment operation data (such as operation parameter data, state data and the like of executing mechanisms such as a heater, a refrigeration compressor, a fan, a humidifier, a dehumidifier and the like). The target reasoning items are data processing tasks to be reasoning, such as reasoning the internal temperature and humidity of the wind cabinet; reasoning the load change of the wind cabinet equipment; reasoning potential faults of the wind cabinet equipment; and (5) reasoning the energy consumption of the wind cabinet. Then, the target item indication information may be a real result of the wind turbine environment monitoring example data in the specified target item, such as when the temperature target item is inferred, the real temperature is the target item indication information of the wind turbine environment monitoring example data.

Step S120, extracting a characterization vector from the first wind turbine environmental monitoring example data according to a first target item network corresponding to the first target reasoning item to obtain a first characterization vector, and extracting a characterization vector from the second wind turbine environmental monitoring example data according to a second target item network corresponding to the second target reasoning item to obtain a second characterization vector.

As one implementation, the first target item network and the second target item network are both single target item networks which are well debugged, for example, the first target item network is a temperature reasoning network, and after temperature-related wind cabinet environment monitoring example data (such as heater data, compressor data, temperature sensor data and the like) are input into the temperature reasoning network, the temperature is deduced.

As an implementation manner, the first characterization vector is an inference link of the first target inference item, feature extraction is performed on the first wind cabinet environment monitoring example data according to the first target item network, and the obtained characterization vector has importance not smaller than an importance threshold, in other words, the first characterization vector is a characterization vector with great influence on target item inference of the first target inference item in the first wind cabinet environment monitoring example data. As an implementation manner, the second characterization vector is an inference link of the second target inference item, and feature extraction is performed on the second wind cabinet environment monitoring example data according to the second target item network, and the obtained characterization vector has importance not smaller than an importance threshold, in other words, the second characterization vector is a characterization vector with great influence on target item inference of the second target inference item in the second wind cabinet environment monitoring example data.

In one embodiment, when the number of features represented by the first token vector and the second token vector is plural, the step of obtaining the first token vector and the second token vector further includes: extracting a characterization vector of the first wind cabinet environment monitoring example data according to a first target item network corresponding to the first target reasoning item to obtain a plurality of first sub-characterization vectors; performing characterization vector combination (such as head-to-tail splicing) on the plurality of first sub-characterization vectors to obtain a first characterization vector; extracting a characterization vector of the second wind cabinet environment monitoring example data according to a second target item network corresponding to the second target reasoning item to obtain a plurality of second sub-characterization vectors; and carrying out characterization vector combination on the plurality of second sub-characterization vectors to obtain a second characterization vector.

Step S130, determining the involvement scores between the first target reasoning matters and the second target reasoning matters according to the involvement between the first characterization vector and the second characterization vector and the involvement between the first target matter indication information and the second target matter indication information.

As an implementation manner, respectively performing low-dimensional mapping processing on the first characterization vector and the second characterization vector, and determining a first mapping characterization vector corresponding to the first characterization vector and a second mapping characterization vector corresponding to the second characterization vector; when the first and second mapping token vectors satisfy a commonality metric requirement (which may be a threshold value indicating a degree of similarity) with each other, a implication score is determined based on the implication between the first and second target item indication information. The low-dimensional mapping process is performed on the feature vector, for example, by performing a hash operation (the mapping token vector at this time is a hash value), or by performing feature embedding (the mapping token vector at this time is an embedding vector). Comparatively, hash operation is adopted on the representation vector, the speed of relativity scoring is faster based on the commonality measurement result of the comparison hash result, but the accuracy is lower than the characteristic embedding, and the specific selection can be determined according to actual needs. Before hash operation, the extracted characterization vectors are combined, so that the data volume of the hash operation is reduced, the operation efficiency is improved, and the real-time performance of task reasoning is facilitated.

In one embodiment, when the difference between the first mapping representation vector and the second mapping representation vector is not greater than the difference threshold, a implication score is determined based on a linear evaluation index (e.g., a product moment correlation coefficient) between the first target item indication information and the second target item indication information, the implication score representing a correlation between the first target item and the second target item, i.e., a value used to evaluate the degree of correlation.

For example, taking the first mapping representation vector and the second mapping representation vector being identical, determining the involvement score as an example according to the linear evaluation index between the first target item indication information and the second target item indication information, to exemplify: the first target inferential matter is implemented as a temperature target inferential matter and the second target inferential matter is implemented as a humidity target inferential matter. Assume that the first wind cabinet environmental monitoring example data includes: example data a (control humidity a), example data B (control humidity B), example data C (control humidity C), wherein the control humidity is target item indication information of the example data, in other words, the target item corresponding to the set of example data is an inference humidity target item. For the group of example data, loading the example data into a single-target item network for reasoning humidity, extracting characterization vectors corresponding to the group of example data respectively, carrying out hash operation on the characterization vectors corresponding to the example data A, the example data B and the example data C respectively, and obtaining a mapping characterization vector A, a mapping characterization vector B and a mapping characterization vector C corresponding to the group of example data respectively. Providing that the second wind cabinet environment monitoring example data comprises example data D (comparison temperature A), example data E (comparison temperature B) and example data F (comparison temperature C), in other words, the target items corresponding to the group of example data are reasoning temperature target items, loading the group of example data into a single target item network of reasoning temperature, extracting characterization vectors corresponding to the group of example data respectively, and carrying out hash operation on the characterization vectors corresponding to the example data D, the example data E and the example data F respectively to obtain a mapping characterization vector D, a mapping characterization vector E and a mapping characterization vector F corresponding to the group of example data respectively.

Then if the indication information corresponding to the example data is highly involved when the mapping token vectors are the same, it indicates that the implication between reasoning humidity target events and reasoning temperature target events is high. If the mapping representation vector A and the mapping representation vector D are the same, and the mapping representation vector C and the mapping representation vector F are the same, then the example data pair (example data A, example data D), (example data C and example data F) for monitoring the environment of the wind cabinet can be determined.

In the embodiment of the application, the involvement score may be calculated by the following formula: s= (Σmn- ΣmΣn/D)/((Σm) ² -(∑m) ² /D)(∑n ² -(∑n) ² /D)) ^-1/2 The method comprises the steps of carrying out a first treatment on the surface of the Wherein D is the number of example data pairs for monitoring the environment of the wind cabinet, m is specifically an indication information value (reference humidity 1 and reference humidity 3) corresponding to the target item of reasoning humidity, and n is specifically an indication information (reference temperature a and reference temperature C) corresponding to the target item of reasoning temperature. Σm is the sum of the control humidity 1 and the control humidity 3, Σn is the sum of the control temperature a and the control temperature C, Σmn is the product of the control humidity 1 and the control temperature a plus the product of the control humidity 3 and the control temperature C.

Step S140, determining a composition architecture of the alternative multi-objective inference network corresponding to the first objective inference item and the second objective inference item when the implication score satisfies the objective item implication condition.

As one embodiment, the inclusion score is determined to satisfy the target event inclusion condition when the absolute value of the inclusion score is not less than the inclusion threshold. As one implementation, the implications between the first and second targeted inference items include positive implications (i.e., positive correlation with each other) and negative implications (i.e., negative correlation with each other), e.g., if the implication score between targeted items 1 and 2 indicates a positive value (e.g., 0.8), then targeted items 1 and 2 are positively involved; if the implication score between target item 1 and target item 2 indicates a negative value (e.g., -0.8), then target item 1 and target item 2 are negatively involved. The composition architecture of the alternative multi-target reasoning network comprises an alternative backbone multi-layer sensor, an alternative first multi-layer sensor corresponding to a first target reasoning item, an alternative first weight distribution module, an alternative second multi-layer sensor corresponding to a second target reasoning item and an alternative second weight distribution module, wherein the alternative backbone multi-layer sensor is used for inputting (i.e. applying) the output characterization vector to the first target reasoning item and the second target reasoning item. The multi-layer sensor is an expert network, and the weight distribution module is a gating network.

For example, calculating the involvement scores between any two of the plurality of target items may determine a adjacency matrix of the involvement scores between all of the target items.

The system is used for a composition architecture schematic of an alternative multi-objective reasoning network, and the alternative multi-objective reasoning network comprises an input layer, a main multi-layer sensor, a first multi-layer sensor, a second multi-layer sensor, a first weight distribution module, a second weight distribution module, a first reasoning module and a second reasoning module. When the target item involvement condition is not satisfied between the first target reasoning item and the second target reasoning item, the target item result can be deduced according to the independent multi-layer perceptron.

As one embodiment, when the absolute value of the implication score between the first target inference item and the second target inference item is less than the implication threshold, it is determined that the target item implication condition is not satisfied between the first target inference item and the second target inference item. As another embodiment, in the composition structure of the alternative multi-objective inference network, the alternative multi-objective inference network may include an input layer, a first multi-layer sensor, a second multi-layer sensor, a first inference module, and a second inference module, where the alternative multi-objective inference network is configured to execute a first objective inference item and a second objective inference item.

In the above two examples, the first embodiment is a composition architecture of the alternative multi-objective inference network when two objective items satisfy the objective item involvement condition, and the second embodiment is a composition architecture of the alternative multi-objective inference network when two objective items do not satisfy the objective item involvement condition, in fact, if the alternative multi-objective inference network is a network for executing more than two objective items, such as objective item 1, objective item 2, and objective item 3, there may be a case in which only one pair of objective items or two pairs of objective items satisfy the objective item involvement condition, taking as an example that objective item 1 and objective item 2 satisfy the objective item involvement condition, objective item 1 and objective item 3 do not satisfy the objective item involvement condition, and objective item 2 and objective item 3 do not satisfy the involvement condition, in this case, in the composition architecture of the alternative multi-objective inference network, objective item 1 corresponds to the first multi-layer sensor, objective item 2 corresponds to the second multi-layer sensor, objective item 3 corresponds to the third multi-layer sensor, and objective item 1 and objective item 2 shares the backbone sensor.

The data reasoning of the alternative multi-objective reasoning network of the first embodiment and the second embodiment is specifically described later.

Further, if a random pair of target items from target item 1, target item 2, and target item 3 all meet the target item implication condition, the composition architecture of the alternative multi-target inference network may include one or all of the following possible scenarios:

1. the alternative multi-objective reasoning network comprises a first multi-layer sensor corresponding to the objective item 1, a second multi-layer sensor corresponding to the objective item 2 and a third multi-layer sensor corresponding to the objective item 3; the target item 1, the target item 2 and the target item 3 also share a backbone multilayer sensor;

2. the alternative multi-objective reasoning network comprises a first multi-layer sensor corresponding to objective item 1, a second multi-layer sensor corresponding to objective item 2 and a third multi-layer sensor corresponding to objective item 3; target item 1 and target item 2 share a backbone multilayer sensor 1, target item 1 and target item 3 share a backbone multilayer sensor 2, and target item 2 and target item 3 share a backbone multilayer sensor 3.

Optionally, after determining the corresponding composition architecture of the alternative multi-objective inference network, the alternative multi-objective inference network needs to be debugged to obtain the multi-objective inference network for multi-data inference.

As one implementation, the alternative multi-objective inference network is debugged according to the debugging data to obtain the multi-objective inference network. The debug data pair comprises debug data forming the data pair and a plurality of comparison target item results, wherein the plurality of comparison target item results are comparison results of the debug data in the plurality of target items. In other words, the plurality of comparison target item results are comparison target item execution results corresponding to the plurality of target items after the plurality of target items (i.e., the plurality of tasks) are completed according to the debug data. For example, taking a plurality of target matters as temperature target reasoning matters (corresponding to first target reasoning matters) and humidity target reasoning matters (corresponding to second target reasoning matters) as examples, the debug data is various sensor data for detecting the temperature and the humidity of the inside and the outside of the wind cabinet, equipment operation data (such as operation parameter data, state data and the like of executing mechanisms such as a heater, a refrigerating compressor, a fan, a humidifier, a dehumidifier and the like) and the like, and the debug data carries the comparison temperature and the comparison humidity. And loading the debugging data into an alternative multi-target inference network, and inferring to obtain an inference temperature and an inference humidity corresponding to the debugging data, and adjusting the alternative multi-target inference network according to the difference between the inference temperature and the comparison temperature and the difference between the comparison humidity and the inference humidity to obtain the multi-target inference network.

Determining a first error (also known as cost, loss) based on a difference between an inference temperature and a reference temperature, in particular when debugging an alternative multi-objective inference network; and determining a second error based on the difference between the comparison humidity and the inference humidity, and optimizing the learnable variables (such as parameters of weight, bias and the like, network layer number, neuron number, learning rate, batch processing size and the like) of the alternative multi-objective inference network along the direction of reducing the fusion result (such as weighted summation) of the first error and the second error.

In summary, according to the alternative constant temperature and humidity air cabinet control method based on the intelligent workshop provided by the embodiment of the application, according to target item indication information respectively corresponding to the plurality of air cabinet environment monitoring example data and the plurality of air cabinet environment monitoring example data, the involvement scores among the plurality of target items are obtained, and when a first target reasoning item and a second target reasoning item in the plurality of target items meet the target item involvement condition, the composition architecture of an alternative multi-target reasoning network is determined. Firstly, the composition architecture of the alternative multi-objective reasoning network is determined under the condition that the involvement of target matters is known, so that the composition architecture is more reasonable, and then, corresponding special modules and general modules are respectively configured for the first target reasoning matters and the second target reasoning matters, so that the alternative multi-objective reasoning network can increase the accuracy of the characterization vector for carrying out target matters reasoning according to the general characterization vector corresponding to the extracted multiple target matters and the special characterization vector corresponding to the single target matters, and further, the accuracy of the multi-objective reasoning network obtained by debugging according to the composition architecture in reasoning the multiple target matters is increased.

In some alternative designs, when determining the first characterization vector or the second characterization vector, the characterization vector extraction obtained by extracting the first wind chest environmental monitoring example data or the second wind chest environmental monitoring example data needs to be selectively determined, so as to obtain a characterization vector with relatively higher importance, so as to increase the accuracy of the obtained target item involvement score. With reference to fig. 3, the corresponding method specifically may include the following steps: step S210, first wind cabinet environment monitoring example data corresponding to a first target reasoning item and second wind cabinet environment monitoring example data corresponding to a second target reasoning item are obtained.

The first wind cabinet environment monitoring example data carries first target item indication information, the second wind cabinet environment monitoring example data carries second target item indication information, and the target item indication information is used for indicating a comparison result of the wind cabinet environment monitoring example data in target items.

Step S220, extracting a characterization vector of the first wind cabinet environment monitoring example data according to a first target item network corresponding to a first target reasoning item to obtain a plurality of first alternative characterization vectors; and determining key token vectors of the plurality of first alternative token vectors to obtain first token vectors.

As an embodiment, the key token vector determination is performed on the plurality of first alternative token vectors, and the manner of obtaining the first token vector is at least one of the following manners, for example:

1. determination by receiver operating characteristic curve (ROC)

The area under the receiver operating characteristic curve (AUC) represents the likelihood that the inference results for positive example data (positive samples) are greater than the inference results for negative example data (negative samples). For example, one example data 1 is arbitrarily determined among all positive example data, one example data 2 is arbitrarily determined among all negative example data, and when two arbitrarily determined example data 1 and 2 are inferred by a classifier, the likelihood (e.g., probability) of reasoning about the example data 1 as a positive example is C ₁ The likelihood of reasoning about example data 2 as a positive example is C ₂ At this time, C1>C2, i.e. the area under the receiver operating characteristic.

As one implementation mode, the first wind cabinet environment monitoring example data comprises P wind cabinet environment monitoring example sub-data, the plurality of first alternative characterization vectors comprise Q first alternative characterization vectors, P is more than or equal to 1, and Q is more than 1. The P wind cabinet environment monitoring example sub-data respectively correspond to the first characterization vectors, and then the process of obtaining the first characterization vectors further includes: reasoning the Q first alternative characterization vectors in the first target reasoning matters to obtain reference target reasoning matters results corresponding to the P wind cabinet environment monitoring example sub-data respectively; determining the area under the receiver operation characteristic curve according to the reference target reasoning item results respectively corresponding to the P wind cabinet environment monitoring example sub-data and the target item indication information respectively corresponding to the P wind cabinet environment monitoring example sub-data; reasoning the characterization vectors except for the T-th characterization vector in the Q first alternative characterization vectors in the first target reasoning matters to obtain T-th target reasoning matter results respectively corresponding to the P wind cabinet environment monitoring example sub-data; determining the area under an S-th receiver operation characteristic curve according to the reference target reasoning item results respectively corresponding to the P wind cabinet environment monitoring example sub-data and target item indication information respectively corresponding to the P wind cabinet environment monitoring example sub-data, wherein T=Q-S, and S is less than or equal to Q; and determining key characterization vectors of the Q first alternative characterization vectors based on the difference value between the area under the S-th receiver operation characteristic curve and the area under the reference receiver operation characteristic curve, so as to obtain first characterization vectors respectively corresponding to the P wind cabinet environment monitoring example sub-data.

For example, if the first target inference item is a temperature target inference item, the P fan cabinet environment monitoring example sub-data are: example data a (control humidity 1), example data B (control humidity 2), example data C (control humidity 3), in other words, the target items corresponding to the set of example data are inferred humidity target items; for the set of example data, loading the set of example data into a single target item network of inference humidity, extracting all characterization vectors (for example, 50 characterization vectors) corresponding to the set of example data respectively, inferring the all characterization vectors to obtain inference humidity 1, inference humidity 2 and inference humidity 3, inputting the comparison humidity 1, the comparison humidity 2, the comparison humidity 3, the inference humidity 1, the inference humidity 2 and the inference humidity 3 into an area under the receiver operation characteristic curve determining module to obtain an area under the receiver operation characteristic curve corresponding to the set of example data, and taking the area under the receiver operation characteristic curve as the area under the reference receiver operation characteristic curve. Then, reducing the characterization vectors corresponding to the example data one by one, for example, reducing the 1 st characterization vector in 50 characterization vectors, reasoning the rest 49 characterization vectors to obtain the reasoning humidity respectively corresponding to the group of example data, and inputting the reasoning humidity respectively corresponding to the group of example data and the contrast humidity respectively corresponding to the group of example data into an area under the receiver operation characteristic curve determining module to obtain the area under the first receiver operation characteristic curve; according to the same thought, reducing the 2 nd characterization vector in the 50 characterization vectors, and reasoning the rest 49 characterization vectors to obtain the area under the operation characteristic curve of the second receiver; until 50 receiver operating characteristic areas were obtained.

The influence (increase or decrease) of the area under the 50 receiver operation characteristic curves on the area under the reference receiver operation characteristic curve is obtained, and the characterization vectors corresponding to the area under the 10 receiver operation characteristic curves, which reduce the area under the reference receiver operation characteristic curve by the maximum, are combined to obtain the characterization vector, namely the target characterization vector corresponding to the set of example data.

2. Determination from a crush stimulus network

As one embodiment, the first target item network corresponding to the first target reasoning item includes a token vector decentration adjustment module, where the token vector decentration adjustment module is configured to determine, during reasoning of the network, an decentration coefficient (i.e. a channel weight) of the extracted token vector. The specific process of obtaining the first characterization vector further includes: inputting a plurality of alternative characterization vectors into a characterization vector eccentric adjustment module, and outputting eccentric coefficients respectively corresponding to the plurality of alternative characterization vectors; and determining key characterization vectors of the plurality of candidate characterization vectors according to the eccentric coefficients respectively corresponding to the plurality of candidate characterization vectors to obtain a first characterization vector.

For example, the first target reasoning item is a temperature target reasoning item, an extrusion excitation module (namely a characterization vector eccentric adjustment module) is added in a temperature reasoning network, Q first alternative characterization vectors corresponding to the wind cabinet environment monitoring example data are extracted according to the temperature reasoning network, eccentric coefficients respectively corresponding to the Q first alternative characterization vectors are calculated according to the extrusion excitation module, and the alternative characterization vector with the largest weight in the Q first alternative characterization vectors is determined as the target characterization vector. Or combining a plurality of candidate token vectors with the largest weights in the Q first candidate token vectors to obtain the target data token vector. Or combining a plurality of characterization vectors with weights not smaller than a threshold value in the Q first alternative characterization vectors to obtain the target data characterization vector.

Step S230, extracting a characterization vector of the second wind cabinet environment monitoring example data according to a second target item network corresponding to a second target reasoning item to obtain a plurality of second alternative characterization vectors; and determining key token vectors of the second alternative token vectors to obtain second token vectors.

Step S240, respectively performing low-dimensional mapping processing on the first characterization vector and the second characterization vector, and determining a first mapping characterization vector corresponding to the first characterization vector and a second mapping characterization vector corresponding to the second characterization vector.

Step S250, determining a implication score according to the implication between the first target item indication information and the second target item indication information when the first mapping representation vector and the second mapping representation vector meet the commonality metric requirement.

In one embodiment, the implication score is determined based on the linear evaluation index between the first target item indication information and the second target item indication information when a difference between the first mapping token vector and the second mapping token vector (e.g., a difference in hash values, where the mapping token vector may be understood as a vector consisting of 0 and 1) is not greater than a difference threshold.

For example, if the first mapping characterization vector and the second mapping characterization vector are the same, the implication score is determined based on a linear assessment indicator between the first target item indication information and the second target item indication information.

Step S260, determining a composition architecture of the alternative multi-objective inference network corresponding to the first objective inference item and the second objective inference item when the implication score satisfies the objective item implication condition.

As one embodiment, the inclusion score is determined to satisfy the target event inclusion condition when the absolute value of the inclusion score is not less than the inclusion threshold.

The composition architecture of the alternative multi-target reasoning network comprises an alternative backbone multi-layer sensor, an alternative first multi-layer sensor corresponding to a first target reasoning item, an alternative first weight distribution module, an alternative second multi-layer sensor corresponding to a second target reasoning item and an alternative second weight distribution module, wherein the alternative backbone multi-layer sensor is used for inputting the obtained characterization vector into the first target reasoning item and the second target reasoning item.

In summary, according to the method provided by the embodiment of the application, the involvement scores among the plurality of target matters are obtained according to the target matter indication information respectively corresponding to the plurality of wind cabinet environment monitoring example data and the plurality of wind cabinet environment monitoring example data, and when the first target reasoning matters and the second target reasoning matters in the plurality of target matters meet the target matter involvement conditions, the composition architecture of the alternative multi-target reasoning network is determined. On one hand, the composition architecture of the alternative multi-objective inference network is determined under the condition that the involvement of objective matters is known, so that the composition architecture is more reasonable; on the other hand, the first target reasoning item and the second target reasoning item are respectively provided with the corresponding special sharing module and the corresponding universal module, so that the alternative multi-target reasoning network can increase the accuracy of the characterization vector for reasoning the target item according to the universal characterization vector corresponding to the plurality of target items and the special characterization vector corresponding to the single target item, and further increase the accuracy of the multi-target reasoning network obtained by debugging according to the composition architecture when reasoning the plurality of target items.

Based on the method, the device and the system, the characterization vectors with high importance are determined to be compared according to the importance of the determined features, so that the operation amount is reduced, the determination speed of the involvement score is improved, the importance determination process is added when the involvement score is determined, the accuracy of the characterization vectors obtained by determination is improved, and the reliability of the involvement score is improved.

After describing the process of determining the composition architecture of the network, the following describes the process of the intelligent workshop-based constant temperature and humidity air cabinet control method provided in the embodiment of the present application, please refer to fig. 4, which specifically includes the following steps: and step S310, acquiring wind cabinet environment monitoring data corresponding to the target constant temperature and humidity wind cabinet, and inputting the wind cabinet environment monitoring data into a multi-target inference network.

The multi-objective reasoning network comprises a main multi-layer perceptron, a first multi-layer perceptron and a first weight distribution module, wherein the main multi-layer perceptron is used for inputting an output characterization vector into the plurality of objective matters.

As one embodiment, the wind turbine environment monitoring data is further used for reasoning among a plurality of target items except the first target reasoning item, for example, the wind turbine environment monitoring data is input into a multi-target reasoning network to obtain a plurality of different data reasoning results, and the plurality of different data reasoning results are respectively corresponding to the plurality of target items.

As an implementation manner, the environmental monitoring data of the wind cabinet records relevant data of the constant temperature and humidity wind cabinet, such as various sensor data for detecting the internal and external humiture of the wind cabinet, and equipment operation data (such as operation parameter data, state data and the like of executing mechanisms such as a heater, a refrigeration compressor, a fan, a humidifier, a dehumidifier and the like).

In one embodiment, the plurality of target items further includes a second target inference item, in other words, the plurality of target items satisfying the target item implication condition with each other includes the first target inference item and the second target inference item satisfying the target item implication condition with each other. Wherein the second target inference item is any one of the plurality of target inference items other than the first target inference item.

In one embodiment, the first target inference result and the second target inference result have a positive (i.e., positive correlation) and negative (i.e., negative correlation) involvement relationship, and if the first target inference result and the second target inference result are positively involved, the value of the involvement score is greater than zero, and at this time, the greater the involvement score, the more relevant the first target inference result and the second target inference result are. If the first and second targeted inferential matters are negatively involved, the implication score is less than zero, at which point the smaller the implication score, the less relevant the first and second targeted inferential matters are.

As one embodiment, the absolute value of the implication score between the first target inference item and the second target inference item is not less than the implication threshold, which is to determine that the first target inference item and the second target inference item satisfy the target item implication condition with respect to each other.

The main multi-layer perceptron is configured to perform characterization vector extraction on the loaded data to obtain a characterization vector for characterizing the commonality of a plurality of target matters; the first multi-layer perceptron is configured to perform token vector extraction on the loaded data to obtain a proprietary token vector that characterizes the first target inference item. As an embodiment, the backbone multi-layer perceptron and the first multi-layer perceptron are feature extraction networks such as affine network layers or convolution network layers, which are not limited in particular.

The first weight distribution module is used for carrying out characterization vector correction (namely, characteristic adjustment) on the characterization vector output by the main multi-layer perceptron to obtain a characterization vector input into the first target reasoning item. The weight distribution module is a gating network. As one implementation mode, the multi-objective reasoning network further comprises a second multi-layer sensor and a second weight distribution module, wherein the second multi-layer sensor corresponds to the second objective reasoning item, and the trunk multi-layer sensor in the multi-objective reasoning network is used for inputting the output characterization vector into the first objective reasoning item and the second objective reasoning item. Optionally, if the plurality of target items has a third target inference item other than the first target inference item and the second target inference item, when an absolute value of the implication score between any two target inference items is not less than the implication threshold value among the first target inference item, the second target inference item, and the third target inference item: the multi-objective reasoning network comprises a first multi-layer sensor and a first weight distribution module corresponding to a first objective reasoning item, a second multi-layer sensor and a second weight distribution module corresponding to a second objective reasoning item, and a third multi-layer sensor and a third weight distribution module corresponding to a third objective reasoning item; the characterization vector output by the main multi-layer perceptron in the multi-target reasoning network is put into the result deduction of the first target reasoning item, the second target reasoning item and the third target reasoning item. Or the multi-objective reasoning network comprises a first multi-layer sensor and a first weight distribution module corresponding to a first objective reasoning item, a second multi-layer sensor and a second weight distribution module corresponding to a second objective reasoning item, and a third multi-layer sensor and a third weight distribution module corresponding to a third objective reasoning item; the main multi-layer perceptron in the multi-objective reasoning network comprises a first universal module, a second universal module and a third universal module, wherein the characterization vector output by the first universal module is input into the result pushing of the first objective reasoning item and the second objective reasoning item, the characterization vector output by the second universal module is input into the result pushing of the first objective reasoning item and the third objective reasoning item, and the characterization vector output by the third universal module is input into the result pushing of the second objective reasoning item and the third objective reasoning item.

Step S320, extracting characterization vectors of the wind cabinet environment monitoring data according to the first multi-layer perceptron and the main multi-layer perceptron to obtain a first characterization vector output by the first multi-layer perceptron and a general characterization vector output by the main multi-layer perceptron.

The first characterization vector characterizes the exclusive characteristic information of the first target reasoning item on the wind cabinet environment monitoring data, and the general characterization vector is used for characterizing the general characteristic information of the first target reasoning item and other target items on the wind cabinet environment monitoring data. As an implementation mode, the multi-target inference network further comprises an input encoding module, the wind cabinet environment monitoring data are input into the multi-target inference network, and the wind cabinet environment monitoring data are encoded according to the input encoding module, so that input characterization vectors corresponding to the wind cabinet environment monitoring data are obtained.

After an input characterization vector corresponding to the wind cabinet environment monitoring data is obtained, the input characterization vector is loaded into a first multi-layer sensor, and the characterization vector is extracted to obtain a first characterization vector; and inputting the input characterization vector into the backbone multi-layer perceptron, and extracting the characterization vector from the input characterization vector to obtain a general characterization vector. Optionally, the wind cabinet environment monitoring data is further used for reasoning in the second target reasoning matters, and the method further includes, while extracting the characterization vector of the wind cabinet environment monitoring data according to the first multi-layer sensor: and extracting the characterization vector of the wind cabinet environment monitoring data according to the second multi-layer perceptron to obtain a second characterization vector output by the second multi-layer perceptron.

The second characterization vector is used for characterizing the exclusive characteristic information of the first target reasoning item on the wind turbine cabinet environment monitoring data; as one embodiment, the universal characterization vector is used for characterizing universal characteristic information of the first target inference item and the second target inference item on the wind cabinet environment monitoring data. As an implementation mode, after the input characterization vector corresponding to the wind cabinet environment monitoring data is obtained, the input characterization vector is input into a second multi-layer sensor, and the characterization vector is extracted to obtain a second characterization vector.

Step S330, the general characterization vector is corrected according to the first weight distribution module, and the corrected characterization vector is integrated with the first characterization vector to obtain a first integrated characterization vector.

Although the general characterization vector characterizes general feature information among a plurality of target matters, the general characterization vector has different influences on target matter reasoning results of the plurality of target matters, and based on the influence, eccentric adjustment can be performed on the general characterization vector through an eccentric coefficient distribution module respectively corresponding to the plurality of target matters, so as to obtain a corrected characterization vector respectively corresponding to the plurality of target matters.

After the input characterization vector corresponding to the wind cabinet environment monitoring data is obtained, the input characterization vector is loaded into the eccentric coefficient distribution module corresponding to the target items respectively. As one embodiment, the method for performing token vector correction on the universal token vector includes: the input characterization vector is input into a first weight distribution module corresponding to the first target reasoning item, a first eccentric coefficient corresponding to the first target reasoning item is determined, eccentric adjustment is carried out on the general characterization vector according to the first eccentric coefficient (the eccentric coefficient is a weight value, and weighting is carried out based on the weight value, namely, eccentric adjustment) to obtain a corrected characterization vector corresponding to the first target reasoning item; and integrating the corrected characterization vector corresponding to the first target reasoning item with the first characterization vector to obtain a first integrated characterization vector.

Optionally, the wind cabinet environment monitoring data is further used for reasoning in the second target reasoning matters, and when the general characterization vector is corrected according to the first weight distribution module, the method further includes: and carrying out characterization vector correction on the universal characterization vector according to the second weight distribution module, and integrating (e.g. adding, splicing or weighting summation) the corrected characterization vector with the second characterization vector to obtain a second integrated characterization vector. As an implementation mode, an input characterization vector is input into a second weight distribution module corresponding to a second target reasoning item, a second eccentric coefficient corresponding to the second target reasoning item is determined, and eccentric adjustment is carried out on the general characterization vector according to the second eccentric coefficient, so that a corrected characterization vector corresponding to the second target reasoning item is obtained; and integrating the corrected characterization vector corresponding to the second target reasoning item with the second characterization vector to obtain a second integrated characterization vector.

Step S340, data reasoning is carried out on the first integrated characterization vector according to a first reasoning module corresponding to the first target reasoning item, and a first reasoning result corresponding to the first target reasoning item is obtained.

The first reasoning module is a network module corresponding to the first target reasoning matters and is configured to infer target matters execution results of the wind turbine environment monitoring data in the first target reasoning matters, and the first reasoning module can be specifically a classifier, such as an affine network layer, a softmax and the like. Optionally, the wind cabinet environment monitoring data is further used for reasoning in the second target reasoning matters, and when the first reasoning module corresponding to the first target reasoning matters performs data reasoning on the first integrated characterization vector, the method further includes: and carrying out data reasoning on the second integrated characterization vector according to a second reasoning module corresponding to the second target reasoning item to obtain a second reasoning result corresponding to the second target reasoning item. The second reasoning module is a network module corresponding to the second target reasoning matters and is configured to infer target matters execution results of the wind cabinet environment monitoring data in the second target reasoning matters.

For example, the first target reasoning item is a temperature target reasoning item in the wind cabinet, the second target reasoning item is a humidity target reasoning item in the wind cabinet, and the first integrated characterization vector is subjected to data reasoning according to a first reasoning module corresponding to the temperature target reasoning item to obtain a reasoning temperature corresponding to the temperature target reasoning item; and carrying out data reasoning on the second integrated characterization vector according to a second reasoning module corresponding to the humidity target reasoning item to obtain the reasoning humidity corresponding to the temperature target reasoning item.

For the alternative multi-target reasoning network, the alternative multi-target reasoning network is debugged to obtain the unchanged composition architecture of the multi-target reasoning network, and the reasoning process after the data is loaded specifically comprises the following steps: loading the wind cabinet environment monitoring data to an input coding module, extracting to obtain an input characterization vector, and respectively inputting the input characterization vector to a main multi-layer sensor, a first multi-layer sensor, a second multi-layer sensor, a first weight distribution module and a second weight distribution module; extracting a general representation vector in an input representation vector according to a main multi-layer perceptron, extracting a first representation vector in the input representation vector according to a first multi-layer perceptron, extracting a second representation vector in the input representation vector according to a second multi-layer perceptron, determining a first weight according to a first weight distribution module, and determining a second weight according to a second weight distribution module; weighting the general characterization vector according to the first weight to obtain a corrected characterization vector corresponding to the first target reasoning item, integrating the corrected characterization vector and the first characterization vector to obtain a first integrated characterization vector, inputting the first integrated characterization vector into the first reasoning module, and outputting a target item execution result of the wind cabinet environment monitoring data on the first target reasoning item; and weighting the universal characterization vector according to the second weight to obtain a corrected characterization vector corresponding to the second target reasoning item, integrating the corrected characterization vector and the second characterization vector to obtain a second integrated characterization vector, and loading the second integrated characterization vector into a second reasoning module to obtain an execution result of the target item of the wind cabinet environment monitoring data on the second target reasoning item.

Optionally, when the target item involvement condition is not satisfied between the first target reasoning item and the second target reasoning item, reasoning can be performed on the target item execution result according to the single multi-layer sensor (i.e. no general multi-layer sensor), specifically, the wind cabinet environment monitoring data is loaded to the input encoding module, the input characterization vector is extracted, and the input characterization vector is respectively input into the first multi-layer sensor and the second multi-layer sensor; extracting a first characterization vector in the input characterization vector according to the first multi-layer perceptron, and extracting a second characterization vector in the input characterization vector according to the second multi-layer perceptron; loading the first characterization vector to a first reasoning module, and outputting a target item execution result of the wind cabinet environment monitoring data on a first target reasoning item; and inputting the second characterization vector into a second reasoning module, and outputting the execution result of the target item of the wind cabinet environment monitoring data on the second target reasoning item.

And step S350, controlling the target constant temperature and humidity air cabinet according to the first reasoning result.

For example, the temperature (for example, the temperature of the target area) obtained by inference is compared based on the set temperature, the heating or cooling mechanism is controlled so that the temperature of the target area reaches the set temperature, the temperature intervention is performed in advance based on the temperature obtained by inference, the temperature change of the target area can be minimized, the effect of constant temperature is achieved, and correspondingly, similar ideas are adopted for controlling the humidity. For the reasoning results corresponding to the target matters such as fault reasoning, load reasoning and the like, corresponding control can be performed according to a preset intervention control strategy, for example, for the fault reasoning results, intervention control such as early warning, shutdown, standby equipment starting and the like is performed according to the importance of the deduced fault equipment and the urgency of fault time, and a specific mode is selected according to actual needs, so that the method is not limited.

According to the constant temperature and constant humidity air cabinet control method based on the intelligent workshop, characterization vector extraction is conducted on air cabinet environment monitoring data according to the first multi-layer perceptron and the main multi-layer perceptron in the multi-objective reasoning network, a first characterization vector and a general characterization vector are obtained, the general characterization vector is corrected according to the first weight distribution module, the corrected characterization vector and the first characterization vector are integrated to obtain a first integrated characterization vector, data reasoning is conducted on the first integrated characterization vector according to the first reasoning module, and a reasoning result corresponding to a first objective reasoning item is obtained. Firstly, a plurality of target items executed in a multi-target reasoning network are target items conforming to the target item involvement conditions, namely, the multi-target reasoning network in the application determines a composition framework under the condition that the target item involvement is clear, so that the multi-target reasoning network obtained by determining the composition framework is more reasonable.

Please refer to fig. 5, which is a schematic structural diagram of a constant temperature and humidity air cabinet control device provided in an embodiment of the present application. The constant temperature and humidity air cabinet control device can be a computer program (including program codes) running in network equipment, for example, the constant temperature and humidity air cabinet control device is an application software; the device can be used for executing corresponding steps in the method provided by the embodiment of the application. As shown in fig. 5, the constant temperature and humidity wind cabinet control apparatus may include: the system comprises a data acquisition module 311, a feature extraction module 312, a feature correction module 313, a data reasoning module 314 and a wind cabinet control module 315.

The data acquisition module 311 is configured to acquire wind cabinet environment monitoring data corresponding to a target constant temperature and humidity wind cabinet, where the wind cabinet environment monitoring data is used to perform data reasoning in a first target reasoning item, input the wind cabinet environment monitoring data into a multi-target reasoning network, where the multi-target reasoning network is configured to execute a plurality of target items including the first target reasoning item, where the plurality of target items meet target item involvement conditions, and the multi-target reasoning network includes a trunk multi-layer sensor, a first multi-layer sensor corresponding to the first target reasoning item, and a first weight distribution module, where the trunk multi-layer sensor is configured to input an output characterization vector into the plurality of target items; the feature extraction module 312 is configured to extract a characterization vector from the wind cabinet environmental monitoring data according to the first multi-layer sensor and the main multi-layer sensor, so as to obtain a first characterization vector output by the first multi-layer sensor and a general characterization vector output by the main multi-layer sensor; the feature correction module 313 is configured to correct the universal token vector according to the first weight distribution module, and integrate the corrected token vector with the first token vector to obtain a first integrated token vector; the data reasoning module 314 is configured to perform data reasoning on the first integrated token vector according to a first reasoning module corresponding to the first target reasoning item, so as to obtain a first reasoning result corresponding to the first target reasoning item; the wind cabinet control module 315 is configured to control the target constant temperature and humidity wind cabinet according to the first reasoning result.

According to one embodiment of the present application, the steps involved in the smart shop-based constant temperature and humidity cabinet control method shown in fig. 4 may be performed by the respective modules in the constant temperature and humidity cabinet control apparatus shown in fig. 5.

According to an embodiment of the present application, each module in the constant temperature and humidity air cabinet control device shown in fig. 5 may be combined into one or several units separately or all, or some (some) of the units may be further split into at least two sub-units with smaller functions, so that the same operation may be implemented without affecting the implementation of the technical effects of the embodiments of the present application. The above modules are divided based on logic functions, and in practical application, the functions of one module may be implemented by at least two units, or the functions of at least two modules may be implemented by one unit. In other embodiments of the present application, the constant temperature and humidity wind cabinet control device may also include other units, and in practical applications, these functions may also be implemented with assistance of other units, and may be implemented by cooperation of at least two units.

According to one embodiment of the present application, the constant temperature and humidity wind cabinet control apparatus as shown in fig. 5 may be constructed by running a computer program (including program code) capable of executing the steps involved in the corresponding method as shown in fig. 4 on a general-purpose computer device such as a computer including a processing component such as a Central Processing Unit (CPU), a random access storage medium (RAM), a read only storage medium (ROM), and the like, and a storage component, and the constant temperature and humidity wind cabinet control method based on a smart shop according to the embodiment of the present application is implemented. The computer program may be recorded on, for example, a computer-readable recording medium, and loaded into and executed by the computing device via the computer-readable recording medium.

Fig. 6 is a schematic structural diagram of a computer device according to an embodiment of the present application. As shown in fig. 6, the above-mentioned computer device 1000 may include: processor 1001, network interface 1004, and memory 1005, and in addition, the above-described computer device 1000 may further include: a user interface 1003, and at least one communication bus 1002. Wherein the communication bus 1002 is used to enable connected communication between these components. The user interface 1003 may include a Display (Display), a Keyboard (Keyboard), and the optional user interface 1003 may further include a standard wired interface, a wireless interface, among others. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a nonvolatile memory (non-volatile memory), such as at least one magnetic disk memory. The memory 1005 may also optionally be at least one storage device remote from the processor 1001. As shown in fig. 6, an operating system, a network communication module, a user interface module, and a device control application may be included in the memory 1005, which is a type of computer-readable storage medium.

In the computer device 1000 shown in FIG. 6, the network interface 1004 may provide network communication functions; while user interface 1003 is primarily used as an interface to provide input; and the processor 1001 may be used to invoke the device control application stored in the memory 1005 to implement the method provided in the above embodiment.

It should be understood that the computer device 1000 described in the embodiment of the present application may perform the description of the method for controlling the constant temperature and humidity air cabinet based on the smart workshop in the embodiment corresponding to fig. 2, and may also perform the description of the device for controlling the constant temperature and humidity air cabinet in the embodiment corresponding to fig. 5, which is not repeated herein. In addition, the description of the beneficial effects of the same method is omitted.

Furthermore, it should be noted here that: the embodiment of the present application further provides a computer readable storage medium, and the computer readable storage medium stores therein a computer program executed by the constant temperature and humidity air cabinet control device mentioned above, and the computer program includes program instructions, when the processor executes the program instructions, the foregoing description of the constant temperature and humidity air cabinet control method based on the smart workshop in the corresponding embodiment of fig. 4 can be executed, and therefore, will not be repeated herein. In addition, the description of the beneficial effects of the same method is omitted. For technical details not disclosed in the embodiments of the computer-readable storage medium according to the present application, please refer to the description of the method embodiments of the present application.

As an example, the above-described program instructions may be executed on one computer device or at least two computer devices disposed at one site, or alternatively, at least two computer devices distributed at least two sites and interconnected by a communication network, which may constitute a blockchain network.

The computer readable storage medium may be the constant temperature and humidity air cabinet control device provided in any one of the foregoing embodiments or a middle storage unit of the foregoing computer device, for example, a hard disk or a memory of the computer device. The computer readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash card (flash card) or the like, which are provided on the computer device. Further, the computer-readable storage medium may also include both a central storage unit and an external storage device of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

The terms first, second and the like in the description and in the claims and drawings of the embodiments of the present application are used for distinguishing between different data or objects and not for describing a particular sequential order. Furthermore, the term "include" and any variations thereof is intended to cover a non-exclusive inclusion. For example, a process, method, apparatus, article, or device that comprises a list of steps or elements is not limited to the list of steps or modules but may, in the alternative, include other steps or modules not listed or inherent to such process, method, apparatus, article, or device.

The embodiment of the present application further provides a computer program product, which includes a computer program/instruction, where the computer program/instruction when executed by a processor implements the description of the foregoing smart workshop-based control method for a constant temperature and humidity air cabinet in the corresponding embodiment of fig. 4, and therefore, a detailed description will not be given here. In addition, the description of the beneficial effects of the same method is omitted. For technical details not disclosed in the embodiments of the computer program product according to the present application, reference is made to the description of the method embodiments of the present application.

Those of ordinary skill in the art will appreciate that the elements and algorithm steps described in connection with the embodiments disclosed herein may be embodied in electronic hardware, in computer software, or in a combination of the two, and that the elements and steps of the examples have been generally described in terms of function in the foregoing description to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

The methods and related devices provided in the embodiments of the present application are described with reference to the method flowcharts and/or structure diagrams provided in the embodiments of the present application, and each flowchart and/or block of the method flowcharts and/or structure diagrams may be implemented by computer program instructions, and combinations of flowcharts and/or blocks in the flowchart and/or block diagrams. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable network connection device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable network connection device, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable network connection device to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or structural diagram block or blocks. These computer program instructions may also be loaded onto a computer or other programmable network connection device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer implemented process such that the instructions which execute on the computer or other programmable device provide steps for implementing the functions specified in the flowchart flow or flows and/or structures. The foregoing disclosure is only illustrative of the preferred embodiments of the present application and is not intended to limit the scope of the claims herein, as the equivalent of the claims herein shall be construed to fall within the scope of the claims herein.

Claims

1. The constant temperature and humidity air cabinet control method based on the intelligent workshop is characterized by being applied to a constant temperature and humidity air cabinet control system, and comprises the following steps: acquiring wind cabinet environment monitoring data corresponding to a target constant temperature and humidity wind cabinet, wherein the wind cabinet environment monitoring data are used for carrying out data reasoning in first target reasoning matters, the wind cabinet environment monitoring data are input into a multi-target reasoning network, the multi-target reasoning network is used for executing a plurality of target matters comprising the first target reasoning matters, the plurality of target matters meet target matter involvement conditions, the multi-target reasoning network comprises a main multi-layer perceptron, a first multi-layer perceptron corresponding to the first target reasoning matters and a first weight distribution module, and the main multi-layer perceptron is used for inputting the output characterization vectors into the plurality of target matters; extracting characterization vectors of the wind cabinet environment monitoring data according to the first multi-layer perceptron and the main multi-layer perceptron to obtain a first characterization vector output by the first multi-layer perceptron and a general characterization vector output by the main multi-layer perceptron; performing characterization vector correction on the general characterization vector according to the first weight distribution module, and integrating the corrected characterization vector with the first characterization vector to obtain a first integrated characterization vector; performing data reasoning on the first integrated characterization vector according to a first reasoning module corresponding to the first target reasoning item to obtain a first reasoning result corresponding to the first target reasoning item; controlling the target constant temperature and humidity air cabinet according to the first reasoning result;

The network architecture of the multi-objective reasoning network is determined by adopting the following steps: acquiring first wind cabinet environment monitoring example data corresponding to first target reasoning matters and second wind cabinet environment monitoring example data corresponding to second target reasoning matters, wherein the first wind cabinet environment monitoring example data carries first target matter indication information, the second wind cabinet environment monitoring example data carries second target matter indication information, and the target matter indication information is used for indicating a comparison result of the wind cabinet environment monitoring example data in the target matters; performing characterization vector extraction on the first wind cabinet environment monitoring example data according to a first target item network corresponding to a first target reasoning item to obtain a first characterization vector, and performing characterization vector extraction on the second wind cabinet environment monitoring example data according to a second target item network corresponding to a second target reasoning item to obtain a second characterization vector; determining a implication score between the first target inference item and the second target inference item based on the implication between the first token vector and the second token vector, and the implication between the first target item indication information and the second target item indication information; determining a composition architecture of an alternative multi-objective inference network corresponding to the first objective inference item and the second objective inference item when the implication score meets a target item implication condition; the composition architecture of the alternative multi-target reasoning network comprises an alternative backbone multi-layer sensor, an alternative first multi-layer sensor and an alternative first weight distribution module corresponding to the first target reasoning item, and an alternative second multi-layer sensor and an alternative second weight distribution module corresponding to the second target reasoning item, wherein the alternative backbone multi-layer sensor is used for inputting the output characterization vector into the first target reasoning item and the second target reasoning item.

2. The method of claim 1, wherein the plurality of target matters comprise a second target reasoning matters, the wind cabinet environment monitoring data is further used for data reasoning in the second target reasoning matters, and the multi-target reasoning network comprises a second multi-layer sensor and a second weight distribution module corresponding to the second target reasoning matters; the method further comprises the steps of: extracting the characterization vector of the wind cabinet environment monitoring data according to the second multi-layer perceptron to obtain a second characterization vector output by the second multi-layer perceptron; performing characterization vector correction on the general characterization vector according to the second weight distribution module, and integrating the corrected characterization vector with the second characterization vector to obtain a second integrated characterization vector; and carrying out data reasoning on the second integrated characterization vector according to a second reasoning module corresponding to the second target reasoning item to obtain a second reasoning result corresponding to the second target reasoning item.

3. The method of claim 2, wherein the plurality of target items includes a temperature target inference item and a humidity target inference item; the data reasoning is performed on the first integrated characterization vector according to a first reasoning module corresponding to the first target reasoning item to obtain a first reasoning result corresponding to the first target reasoning item, including: performing data reasoning on the first integrated characterization vector according to a first reasoning module corresponding to the temperature target reasoning item to obtain a reasoning temperature corresponding to the temperature target reasoning item; the data reasoning is performed on the second integrated characterization vector according to a second reasoning module corresponding to the second target reasoning item to obtain a second reasoning result corresponding to the second target reasoning item, including: and carrying out data reasoning on the second integrated characterization vector according to a second reasoning module corresponding to the humidity target reasoning item to obtain the reasoning humidity corresponding to the humidity target reasoning item.

4. The method of claim 1, wherein the determining a involvement score between the first target inference item and the second target inference item based on the involvement between the first characterization vector and the second characterization vector and the involvement between the first target item indication information and the second target item indication information comprises: respectively carrying out low-dimensional mapping processing on the first characterization vector and the second characterization vector, and determining a first mapping characterization vector corresponding to the first characterization vector and a second mapping characterization vector corresponding to the second characterization vector; determining the implication score according to the implication between the first and second target item indication information when the first and second mapping token vectors meet a commonality metric requirement with each other.

5. The method of claim 3, wherein the extracting the characterization vector from the first target item network corresponding to the first target inference item to obtain a first characterization vector, and the extracting the characterization vector from the second target item network corresponding to the second target inference item to obtain a second characterization vector from the second wind cabinet environment monitoring example data, comprises: extracting a characterization vector of the first wind cabinet environment monitoring example data according to a first target item network corresponding to the first target reasoning item to obtain a plurality of first sub-characterization vectors; carrying out characterization vector combination on the plurality of first sub-characterization vectors to obtain a first characterization vector; extracting a characterization vector of the second wind cabinet environment monitoring example data according to a second target item network corresponding to the second target reasoning item to obtain a plurality of second sub-characterization vectors; and carrying out characterization vector combination on the plurality of second sub-characterization vectors to obtain a second characterization vector.

6. The method of claim 1, wherein the extracting the characterization vector from the first target item network corresponding to the first target inference item to obtain a first characterization vector, and the extracting the characterization vector from the second target item network corresponding to the second target inference item to obtain a second characterization vector from the second wind cabinet environment monitoring example data, comprises: extracting the characterization vector of the first wind cabinet environment monitoring example data according to a first target item network corresponding to the first target reasoning item to obtain a plurality of first alternative characterization vectors; performing key characterization vector determination on the plurality of first alternative characterization vectors to obtain the first characterization vectors; extracting the characterization vector of the second wind cabinet environment monitoring example data according to a second target item network corresponding to the second target reasoning item to obtain a plurality of second alternative characterization vectors; and determining key characterization vectors of the second alternative characterization vectors to obtain the second characterization vectors.

7. The method of claim 6, wherein the first cabinet environmental monitoring example data includes P cabinet environmental monitoring example sub-data, the plurality of first alternative characterization vectors includes Q first alternative characterization vectors, P is greater than or equal to 1, Q is greater than 1; performing key token vector determination on the plurality of first alternative token vectors to obtain the first token vectors, including: reasoning the Q first alternative characterization vectors in the first target reasoning matters to obtain reference target reasoning matters results corresponding to the P wind cabinet environment monitoring example sub-data respectively; determining the area under the receiver operation characteristic curve according to the reference target reasoning item results respectively corresponding to the P wind cabinet environment monitoring example sub-data and the target item indication information respectively corresponding to the P wind cabinet environment monitoring example sub-data; reasoning the characterization vectors except for the T-th alternative characterization vector in the Q first alternative characterization vectors in the first target reasoning matters to obtain T-th target reasoning matters corresponding to the P wind cabinet environment monitoring example sub-data respectively; determining the area under an S-th receiver operation characteristic curve according to the reference target reasoning item results respectively corresponding to the P wind cabinet environment monitoring example sub-data and target item indication information respectively corresponding to the P wind cabinet environment monitoring example sub-data, wherein T=Q-S, and S is less than or equal to Q; and determining the key characterization vectors of the Q first alternative characterization vectors according to the difference value between the area under the S-th receiver operation characteristic curve and the area under the reference receiver operation characteristic curve, so as to obtain the first characterization vectors respectively corresponding to the P wind cabinet environment monitoring example sub-data.

8. The method according to claim 6, wherein the first target item network corresponding to the first target reasoning item includes a characterization vector decentration adjustment module, and the characterization vector decentration adjustment module is configured to determine an decentration coefficient of the extracted characterization vector in a reasoning process of the network; performing key token vector determination on the plurality of first alternative token vectors to obtain the first token vectors, including: inputting the plurality of first alternative characterization vectors into the characterization vector eccentric adjustment module, and outputting eccentric coefficients respectively corresponding to the plurality of first alternative characterization vectors; and determining key characterization vectors of the plurality of first alternative characterization vectors according to the eccentric coefficients respectively corresponding to the plurality of first alternative characterization vectors to obtain the first characterization vectors.

9. A constant temperature and humidity air cabinet control system, characterized by comprising: a processor; and a memory, wherein the memory has stored therein a computer program which, when executed by the processor, causes the processor to perform the method of any of claims 1-8.