Power equipment quality evaluation method of double-layer feedback architecture
Through the power equipment quality evaluation method with a double-layer feedback architecture, the deep belief network and BP neural network are used to solve the problem of reliance on subjective judgment in the existing technology, achieve higher evaluation accuracy and precision, simplify the complex evaluation process, and improve the reliability of power equipment quality evaluation.
Patent Information
- Application Number
- CN202510922759.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-07-09
- Filing Date
- 2025-07-04
- Publication Date
- 2025-09-19
AI Technical Summary
The existing power equipment quality evaluation method relies too much on subjective judgment. As the number of evaluation indicators increases, the consistency is difficult to verify, resulting in a decrease in the accuracy of the evaluation results, and fails to fully explore the intrinsic connection between the evaluation indicators and equipment quality.
A two-layer feedback architecture, including deep belief network and BP neural network, is adopted to construct a power equipment quality evaluation index system. The quality characteristics of power equipment are extracted through the deep belief network, and the BP neural network is used to perform quality grade classification and score prediction. The feedback mechanism is introduced into two sub-processes: classification and evaluation, which reduces the influence of subjective judgment and improves the evaluation accuracy.
Through the design of a two-layer feedback architecture, the evaluation complexity is reduced, the accuracy and precision of power equipment quality evaluation are improved, accurate judgments can be made when facing complex problems, the BP neural network training of each sub-process is simplified, and the output accuracy and reliability of the model are improved.
Smart Images
Figure CN120671558A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of smart grids, and in particular relates to a method for evaluating the quality of power equipment with a double-layer feedback architecture. Background Art
[0002] Power equipment is a core asset of power grid companies. Its quality level is affected by different operating conditions and maintenance levels, directly impacting the reliability and economic viability of the power grid. Therefore, evaluating the quality of power equipment is crucial. However, when using traditional evaluation methods such as the Analytic Hierarchy Process (AHP), the increase in evaluation indicators leads to increased complexity in the judgment matrix, which also increases the difficulty of consistency verification and reduces the reliability of the results. Existing traditional quality evaluation methods rely too much on subjective judgment and ignore the inherent connection between power equipment operating data and equipment quality. A power equipment quality evaluation method is needed that can address complex issues, reduce the impact of subjective judgment, and fully explore the inherent connection between evaluation indicators and quality levels. Summary of the Invention
[0003] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a power equipment quality evaluation method with a double-layer feedback architecture, which solves the technical problems that the current quality evaluation method relies too much on subjective judgment, is difficult to verify consistency as the number of indicators increases, and has reduced accuracy.
[0004] The purpose of the present invention is achieved as follows: a method for evaluating the quality of power equipment with a double-layer feedback architecture, which includes: S1, constructing a double-layer feedback architecture for evaluating the quality of power equipment; S2, constructing a power equipment quality evaluation index system; S3, obtaining power equipment data based on the power equipment quality evaluation index system, and sending the data into the double-layer feedback architecture to extract the power equipment quality characteristic value, and as the first layer output; S4, the power equipment quality characteristics output by the first layer of the double-layer feedback architecture are used as the input of the quality grade classification BP neural network in the second layer, and the quality grade classification BP neural network outputs the quality grade to which the power equipment quality belongs; S5, the quality grade to which the equipment quality belongs is fed back to the quality score BP neural network selection unit, and the selection unit selects the quality score BP neural network corresponding to the quality grade in the second layer according to the fed-back quality grade, and inputs the power equipment quality characteristic value output by the first layer into the score BP neural network to obtain the power equipment quality score.
[0005] Furthermore, the two-layer feedback architecture includes a first layer and a second layer; the first layer is a deep belief network; the second layer includes a quality grade classification BP neural network, a quality score BP neural network selection unit and several quality score BP neural networks.
[0006] Furthermore, the network of the first layer is composed of a multi-layer restricted Boltzmann machine, and the number of neurons in its input layer is equal to the number of evaluation indicators; the number of neurons in the input layer of the quality grade classification BP neural network of the second layer is equal to the number of quality characteristics of the power equipment output by the first layer, and the number of neurons in the output layer is equal to the number of quality grades of the power equipment.
[0007] Furthermore, the number of quality score BP neural networks in the second layer is equal to the number of quality grades of power equipment; the number of input layer neurons of the quality score BP neural network is equal to the number of quality features of power equipment output by the first layer, and the number of output layer neurons of the quality score BP neural network is 1.
[0008] Furthermore, the construction of the power equipment quality evaluation index system includes the following steps: the first step is based on the observability analysis of complex physical systems, and the power equipment quality evaluation indicators include general quality evaluation indicators of power equipment and special quality evaluation indicators of similar power equipment; the second step is to select general quality evaluation indicators of power equipment and special quality evaluation indicators of similar power equipment on the premise of running through the entire life cycle of power equipment and with the goal of fully reflecting the quality level of power equipment, and obtain relevant data based on the selected indicators and perform forward and normalization processing.
[0009] Furthermore, the last layer activation function of the quality grade classification BP neural network adopts a softmax function, and when the loss function reaches a set value, the activation function value quality grade is output.
[0010] Beneficial effects of the present invention:
[0011] The deep features of evaluation indicators are extracted through deep belief networks, and quality grade classification and quality level evaluation are performed through BP neural networks. This can not only reduce the impact of subjective judgment but also make accurate judgments when faced with complex evaluations.
[0012] The first layer of the two-layer feedback architecture uses a deep belief network to achieve noise reduction and dimensionality reduction of the original data of power equipment quality evaluation. The obtained low-dimensional power equipment quality characteristic values can reduce the complexity of the two-layer neural network and improve its output accuracy; the second layer of the two-layer feedback architecture introduces a feedback mechanism, which divides the power equipment evaluation into two sub-processes: classification and evaluation, simplifying the complexity of the BP neural network in each sub-process, overcoming the problem that the network training parameters are difficult to converge when a single BP neural network evaluates equipment of different quality levels at the same time, and improving the accuracy of power equipment quality evaluation. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 It is a schematic diagram of the first-level power equipment quality evaluation method of the present invention;
[0014] Figure 2 is a schematic diagram of the second-level power equipment quality evaluation method of the present invention;
[0015] Figure 3 This is a comparison diagram of the quality grade quality score BP neural network output device quality score and the actual quality score of the present invention. DETAILED DESCRIPTION
[0016] The present invention will be further described in detail below with reference to the accompanying drawings. It should be pointed out that all directional words such as up, down, front, back, left, and right appearing in the present invention do not limit the present invention, but are only for the purpose of more clearly illustrating and explaining the present invention.
[0017] Example 1
[0018] like Figure 1-3 As shown, this embodiment discloses a method for evaluating the quality of power equipment using a double-layer feedback architecture, which specifically includes the following steps:
[0019] First, a two-layer feedback architecture for power equipment quality evaluation is constructed, specifically comprising the following steps: the first layer of the two-layer feedback architecture is a deep belief network, the number of neurons in its input layer being equal to the number of evaluation indicators of the quality evaluation system in claim 3, and the number of neurons in its output layer being equal to the number of quality characteristics of the power equipment. The second layer of the two-layer feedback architecture comprises: a quality grade classification BP neural network, the number of neurons in its input layer being equal to the number of quality characteristics of the power equipment output by the first layer of the two-layer feedback architecture, and the number of neurons in its output layer being equal to the number of quality grades of the power equipment; a quality score BP neural network selection unit; the number of quality score BP neural networks being equal to the number of quality grades of the power equipment, the number of neurons in their input layer being equal to the number of quality characteristics of the power equipment output by the first layer of the two-layer feedback architecture, and the number of neurons in their output layer being one.
[0020] Secondly, a power equipment quality evaluation index system is constructed, which specifically includes the following steps: based on the observability analysis of complex physical systems, the power equipment quality evaluation indicators include general quality evaluation indicators for power equipment and special quality evaluation indicators for similar power equipment; with the entire life cycle of power equipment as the premise and the goal of fully reflecting the quality level of power equipment, general quality evaluation indicators for power equipment and special quality evaluation indicators for similar power equipment are selected, and relevant data are obtained based on the selected indicators, and then the obtained equipment-related data are normalized and positively processed; and the quality of power equipment is divided into five quality levels according to the historical quality scores of power equipment.
[0021] Then, the processed data is sent to the first-layer deep belief network of the double-layer feedback architecture described in step S1, and the restricted Boltzmann machine is trained layer by layer through the contrast divergence algorithm to gradually extract the characteristic values of the power equipment as the first-layer output; the quality characteristics of the power equipment output of the first layer of the double-layer feedback architecture are only used as the input of the quality grade classification BP neural network in the second layer, and the softmax function is used as the activation function of the last layer of the quality grade classification BP neural network. When the loss function reaches the set value, the quality grade with the highest activation function value is used as the output.
[0022] Finally, the quality grade of the power equipment is fed back to the quality score BP neural network selection unit. The selection unit selects the quality score BP neural network corresponding to the quality grade in the second layer according to the fed-back quality grade, and inputs the power equipment quality feature value output by the first layer in step S3 into the score BP neural network to obtain the quality score of the power equipment.
[0023] Example 2
[0024] like Figure 1-3 As shown, this embodiment discloses a method for evaluating the quality of power equipment using a dual-layer feedback architecture, comprising: constructing a dual-layer feedback architecture for evaluating the quality of power equipment; the first layer of the dual-layer feedback architecture comprises a deep belief network formed by eight restricted Boltzmann machines (RBMs); the number of neurons in the input layer equals the number of evaluation indicators; and the number of neurons in the output layer equals the number of quality features of the power equipment. The second layer of the dual-layer feedback architecture comprises a quality grade classification BP neural network; the number of neurons in the input layer equals the number of quality features of the power equipment output by the first layer of the dual-layer feedback architecture; and the number of neurons in the output layer equals the number of quality grades of the power equipment; and a quality score BP neural network selection unit; the number of quality score BP neural networks equals the number of quality grades of the power equipment; the number of neurons in the input layer equals the number of quality features of the power equipment output by the first layer of the dual-layer feedback architecture; and the number of neurons in the output layer is 1.
[0025] Constructing a power equipment quality evaluation index system based on the observability analysis of complex physical systems, the power equipment quality evaluation index includes general quality evaluation indexes for power equipment and specific quality evaluation indexes for similar power equipment. With the premise of covering the entire life cycle of power equipment and the goal of fully reflecting the quality level of power equipment, general quality evaluation indexes for power equipment and specific quality evaluation indexes for similar power equipment are selected, and relevant data is obtained based on the selected indicators. The obtained equipment-related data is then normalized and positively processed according to the following formula:
[0026]
[0027] in, represents the normalized positive correlation data, represents the reverse correlation data after normalization, Z i Represents the relevant data of the i-th category of power equipment, Z i,max Represents the maximum value in the i-th category related data, Z i,min Represents the minimum value in the i-th category related data.
[0028] Then, the quality of power equipment is divided into five quality levels according to the historical quality scores of power equipment. The relationship between the equipment quality level and the equipment score is shown in the following table:
[0029] Equipment quality level Equipment quality score healthy 100～88 good 88～76 generally 76～64 Poor 64～52 Danger 52～40
[0030] The processed data is fed into the first-layer deep belief network of the two-layer feedback architecture described in step S1 to extract the quality features of the power equipment as the first-layer output. The specific implementation method is as follows: the processed power equipment-related data is fed into the first-layer restricted Boltzmann machine (RBM) and used as the visible layer vector v1 to calculate the hidden layer activation probability using the following formula:
[0031]
[0032] Where, P(h 1,j =1|v1) represents the activation probability of the jth neuron in the hidden layer of the restricted Boltzmann machine (RBM) when the visible layer of the restricted Boltzmann machine (RBM) is known, h 1,j represents the jth neuron in the hidden layer, v1 is the visible layer vector composed of the processed power equipment related data, and v 1,i represents the i-th neuron in the visible layer, b 1,j is the bias of the jth neuron in the hidden layer, w ij is the connection weight between the i-th neuron in the visible layer and the j-th neuron in the hidden layer, σ is the sigmoid activation function, where e -x Will Transform to nonlinear space.
[0033] According to the hidden layer activation probability, random sampling is performed to determine the h of each neuron in the hidden layer j After the activation state, the hidden layer vector h1 is obtained and the activation probability of the reconstructed visible layer is calculated by the following formula:
[0034]
[0035] In the formula, P(v 2,i =1|h1) represents the activation probability of the i-th neuron in the reconstructed visible layer v2 of the restricted Boltzmann machine (RBM) when the hidden layer h1 of the restricted Boltzmann machine (RBM) is known, v 2,i Represents the reconstruction of the i-th neuron in the visible layer, h1 is the hidden layer vector, a1,i is the bias of the i-th neuron in the visible layer, h 1,j represents the jth neuron in the hidden layer, w ij is the connection weight between the i-th neuron in the visible layer and the j-th neuron in the hidden layer, σ is the sigmoid activation function, where e -x Will Transform to nonlinear space.
[0036] According to the activation probability of the visible layer, random sampling is performed to determine the reconstructed neurons v of the visible layer 2,i After the activation state, the visible layer reconstruction vector v2 is obtained and the reconstructed hidden layer activation probability is calculated by the following formula:
[0037]
[0038] Where, P(h 2,j =1|v2) represents the activation probability of the jth neuron in the reconstructed hidden layer h2 of the restricted Boltzmann machine (RBM) when the RBM reconstructed visible layer is known, h 2,j represents the jth neuron of the reconstructed hidden layer, v2 is the reconstructed visible layer vector, b 2,j To reconstruct the bias of the jth neuron in the hidden layer, w ij is the connection weight between the i-th neuron in the visible layer and the j-th neuron in the hidden layer, σ is the sigmoid activation function, where e -x Will Transform to nonlinear space.
[0039] According to the activation probability of the reconstructed hidden layer, random sampling is performed to determine the h of each neuron in the reconstructed hidden layer. 2,j After the activation state, the hidden layer vector h2 is obtained; the correction amount of weight and bias is calculated according to the following formula:
[0040]
[0041] Among them, α represents the learning rate, w (t) represents the weight after training the restricted Boltzmann machine (RBM) t times, a (t) represents the visible layer bias after t times of training of restricted Boltzmann machine (RBM), b (t) represents the hidden layer bias after the restricted Boltzmann machine (RBM) is trained t times, dw (t) Represents the weight correction amount when the restricted Boltzmann machine (RBM) is trained t times, da (t) Represents the visible layer bias correction when the restricted Boltzmann machine (RBM) is trained t times, db (t) It represents the hidden layer bias correction when the restricted Boltzmann machine (RBM) is trained t times.
[0042] After reaching the required number of training times, the training is stopped and the deep quality features of the power equipment are output; further, the quality features of the power equipment output by the first layer of the two-layer feedback architecture are only used as the input of the quality grade classification BP neural network in the second layer, and the softmax function is used as the activation function of the last layer of the quality grade classification BP neural network. When the loss function reaches the set value, the quality grade with the highest activation function value is used as the output.
[0043] Finally, the quality grade of the power equipment is fed back to the quality score BP neural network selection unit. The selection unit selects the quality score BP neural network corresponding to the quality grade in the second layer according to the fed-back quality grade, and inputs the power equipment quality feature value output by the first layer in step S3 into the score BP neural network to obtain the power equipment quality score. The quality score of the equipment output by the quality score BP neural network corresponding to the good quality grade is compared with the actual quality score (such as Figure 3 shown).
[0044] The two-layer feedback architecture divides the quality evaluation of power equipment into two stages: feature extraction and quality grade classification / quality score prediction. It can not only better reflect the inherent structure and regularity of the data, making the classification / prediction results of the model more convincing and credible, but also effectively process complex data. When faced with complex problems, it reduces the complexity of the model and improves the speed and accuracy of quality grade / quality score prediction.
[0045] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes based on the technical solutions and concepts of the present invention within the technical scope disclosed by the present invention, and they should be covered by the scope of protection of the present invention.
Claims
1. A method for evaluating the quality of power equipment using a double-layer feedback architecture, characterized in that: include: S1. Construct a two-layer feedback architecture for power equipment quality evaluation; S2. Construct a quality evaluation index system for power equipment; S3. Acquire power equipment data according to the power equipment quality evaluation index system, and input the data into the two-layer feedback architecture to extract power equipment quality feature values as the first-layer output; S4. The power equipment quality features output by the first layer of the two-layer feedback architecture are used as input to the quality grade classification BP neural network in the second layer, and the quality grade classification BP neural network outputs the quality grade of the power equipment; S5. The quality grade of the equipment quality is fed back to the quality score BP neural network selection unit. The selection unit selects the quality score BP neural network corresponding to the quality grade in the second layer according to the fed-back quality grade, and inputs the power equipment quality feature value output by the first layer into the score BP neural network to obtain the power equipment quality score.
2. The power equipment quality evaluation method with a double-layer feedback architecture according to claim 1, characterized in that: The two-layer feedback architecture includes a first layer and a second layer; the first layer is a deep belief network; the second layer includes a quality grade classification BP neural network, a quality score BP neural network selection unit and several quality score BP neural networks.
3. The power equipment quality evaluation method with a double-layer feedback architecture according to claim 2, characterized in that: The network of the first layer is composed of a multi-layer restricted Boltzmann machine, and the number of neurons in its input layer is equal to the number of evaluation indicators; the quality grade classification BP neural network of the second layer has the number of neurons in its input layer equal to the number of power equipment quality features output by the first layer, and the number of neurons in its output layer is equal to the number of quality grades of the power equipment.
4. The power equipment quality evaluation method with a double-layer feedback architecture according to claim 3 is characterized by: The number of quality score BP neural networks in the second layer is equal to the number of quality grades of power equipment; the number of input layer neurons of the quality score BP neural network is equal to the number of quality features of power equipment output by the first layer, and the number of output layer neurons of the quality score BP neural network is 1.
5. The power equipment quality evaluation method with a double-layer feedback architecture according to claim 1, characterized in that: The construction of the power equipment quality evaluation index system includes the following steps: The first step is to analyze the observability of complex physical systems. The quality evaluation indicators of power equipment include general quality evaluation indicators of power equipment and special quality evaluation indicators of similar power equipment. The second step is to select general quality evaluation indicators for power equipment and special quality evaluation indicators for similar power equipment, based on the premise of covering the entire life cycle of power equipment and aiming to fully reflect the quality level of power equipment. Relevant data are obtained based on the selected indicators and are processed positively and normalized.
6. The power equipment quality evaluation method with a double-layer feedback architecture according to claim 1, characterized in that: The last layer activation function of the quality grade classification BP neural network adopts the softmax function, and when the loss function reaches the set value, the activation function value quality grade is output.