Communication architecture of power monitoring system based on incidence matrix model

The three-tier business model is mainly applicable to the existing 2M dedicated line business. The models of these services include fibre optic cable layer, transport layer (SDH), service layer, etc., from the bottom-up. To better understand the three-layer model, let’s take relay protection of a particular power company as an example. The schematic diagram of the three-layer business model of relay protection is shown in Figure 2.

Fig. 2

The variable relay protection business is equipped with two independent paths, active and standby, at the bottom layer’s network layer and optical cable layer. One of the shorter orange paths is the active path, and the yellow path is the backup path. Interconnected alarms are generated when the main path of the service model, the optical cable layer, the network layer equipment, or the link carried by the backup path fails [6]. However, current alarms mainly come from the failure of information such as transmission equipment ports and multiplex sections. This does not consider the potential impact on the business when the alternate path is interrupted.

The four-layer service model is mainly applicable to the existing system protection services carried on the data network, for example, panoramic monitoring, dispatch automation, WAMS, etc. These business models include fibre optic cable layer, transmission layer (SDH), data layer (dispatch data network), business layer, etc., from the bottom-up. To better understand the three-tier model, we show a schematic diagram of the four-tier business model by analysing a particular substation dispatching automation business as an example. In Figure 3, the dispatch automation service is configured with three paths: primary, standby, and detour on the optical cable layer, SDH layer, and data layer of the bottom layer.

Fig. 3

Based on the above two load-bearing models, we need to build an association model from the underlying fault to the business risk. At present, the correlation analysis between equipment failures and alarms and the correlation analysis methods between alarms have been thoroughly studied. However, in the field of electric power communication, there is not enough research on the influence of fault alarm and its inducement, as well as the mapping relationship between fault and business [7]. So then, we have built the alarm-fault correlation analysis model of the electric power communication network.

We assume that the total number of factors is M. The type of factor is n_f. Different kinds of factors constitute a set F = {F_k}. There is at |F| = n_f this time. The factor type F_k is composed of h_k factors. Then the factor set has the following relationship: (1) ∑Fk∈F|Fk|=M \sum\nolimits_{{F_k} \in F} |{F_k}| = M When classifying, we need to ensure that the various types of factors are independent of each other. There are mutually exclusive and compatible relations among factors in the same kind of factor set [8]. The fault alarm of grade j is marked as G_j. Through historical data and empirical analysis, we can get the set F_G of the types of factors that may lead to it. If a factor i may cause a fault alarm G_j, set its weight x_{i j} to 1, otherwise set its weight to 0. We assume that the probability of occurrence of a fault alarm G_j is PjG(t) P_j^G(t) . In the factor typeset F_G, the trigger probability of the factor type k occurring is PkjF(t) P_{kj}^F(t) . The probability of occurrence of failure G_j is: (2) PjG(t)=1−∏Fk∈FG(1−PkjF(t)) P_j^G(t) = 1 - \prod\limits_{{F_k} \in {F_G}} (1 - P_{kj}^F(t)) Since the relevance of fault alarms due to various causes is not sure, we analyse the quantification process of PkjF(t) P_{kj}^F(t) . We assume that the weight of each factor i is w_i. The weight set is W = {w_i}. Considering the repulsion and compatibility of different factors, we can obtain the relationship between PkjF(t) P_{kj}^F(t) and each important event as follows: (3) PkjF(t)=Σi∈FkwixijPi(t)−Σi<l=2,i,l∈FkwiwlxijxljPil(t)+Σi<l<r=3,i,l,r∈FkwiwlwrxijxljxrjPilr(t)⋯+(−1)hk−1Πi∈FkwixiP12⋯hk(t) \matrix{ {P_{kj}^F(t) = {\Sigma _{i \in {F_k}}}{w_i}{x_{ij}}{P_i}(t) - {\Sigma _{i < l = 2,i,l \in {F_k}}}{w_i}{w_l}{x_{ij}}{x_{lj}}{P_{il}}(t) + } \hfill \cr {{\Sigma _{i < l < r = 3,i,l,r \in {F_k}}}{w_i}{w_l}{w_r}{x_{ij}}{x_{lj}}{x_{rj}}{P_{ilr}}(t) \cdots + {{( - 1)}^{{h_k} - 1}}\mathop \Pi \limits_{i \in {F_k}} {w_i}{x_i}{P_{12 \cdots {h_k}}}(t)} \hfill \cr } Among them, P_{i j…l}(t) represents the inducement probability of a factor i, j, …, l occurring at the same time. We need to consider that the weights should maximise the inducement probability PkjF(t) P_{kj}^F(t) of k factors. The above formula needs to meet the following optimisation goals: (4) minZ1(W)=PkjF(t) \min {Z_1}(W) = P_{kj}^F(t) For uncertain systems, there is an effective way to quantify information entropy [9]. To normalize the inducement probability and dependence degree of various factors F_k, this paper proposes the following constraints: (5) ∀Fk,−Σi∈Fkwixij∂PkjF(t)∂Pi(t)=1 \forall {F_k}, - {\Sigma _{i \in {F_k}}}{w_i}{x_{ij}}{{\partial P_{kj}^F(t)} \over {\partial {P_i}(t)}} = 1 Among them ∂PkjF(t)∂Pi(t) {{\partial P_{kj}^F(t)} \over {\partial {P_i}(t)}} is the inducement probability of k factors. PkjF(t) P_{kj}^F(t) is the partial derivative of the trigger probability P_i(t) concerning the factor i. This reflects the dependence of the k type inducement set on the factor i. (6) minZ2(W)=−Σi∈Fkwixij∂PkjF(t)∂Pi(t)log2(wixij∂PkjF(t)∂Pi(t)) \min {Z_2}(W) = - {\Sigma _{i \in {F_k}}}{w_i}{x_{ij}}{{\partial P_{kj}^F(t)} \over {\partial {P_i}(t)}}\mathop {\log }\nolimits_2 \left( {{w_i}{x_{ij}}{{\partial P_{kj}^F(t)} \over {\partial {P_i}(t)}}} \right) Taking into account the above two goals, the final optimisation goals and corresponding constraints are defined as follows: (7) minZ(W)=ηZ1(W)+(1−η)Z2(W)s.t.{ ∑i∈Fkwixij∂PkjF(t)∂Pi(t)0<wi<1 \matrix{ {\min Z(W) = \eta {Z_1}(W) + (1 - \eta ){Z_2}(W)} \hfill \cr {s.t.\left\{ {\matrix{ {\sum\nolimits_{i \in {F_k}} {w_i}{x_{ij}}{{\partial P_{kj}^F(t)} \over {\partial {P_i}(t)}}} \hfill \cr {0 < {w_i} < 1} \hfill \cr } } \right.} \hfill \cr } Among them η is the scale factor of entropy weight and satisfies 0 < η < 1. In the constraints, we require the weight w_i of each factor i in the factor category k to be between 0 and 1.

2.3.2

The extended Newton iteration method

In the optimisation model, we can prove that the problem is a continuous optimisation problem. Since the goal and constraints of the problem are both quadratic, the practical solution method is the Lagrange multiplier method [10]. To simplify the solution method, this paper proposes an extended Newton iteration method to solve the problem. The process is shown in Figure 4. The specific process is as follows:

Step 1: We set the initial solution as W0=(1hk,⋯,1hk) {W_0} = \left( {\sqrt {{1 \over {{h_k}}}} , \cdots ,\sqrt {{1 \over {{h_k}}}} } \right) . The initial value is Z₀ = +∞. The feasible solution is W′ = ∅. The initial number of iterations is n_c = 0. Then we go to step 2.

Step 2: We let W1=W0−Z(W0)∂Z(W0)∂W0,nc=nc+1 {W_1} = {W_0} - {{Z({W_0})} \over {{{\partial Z({W_0})} \over {\partial {W_0}}}}},{n_c} = {n_c} + 1 , n_c = n_c + 1 go to step 3.

Step 3: Normalise W₁ according to the constraints. Let us assume W1=W1∑i∈Fkwi2 {W_1} = {{{W_1}} \over {\sum\nolimits_{i \in {F_k}} w_i^2}} . Calculate B. If Z(W₁) < Z₀ then let Z₀ = Z(W₁), W′ = W₁. Then we go to step 4;

Step 4: If |Z(W₁) − Z(W₀)| ≤ ε, output the optimal value Z_o = min {Z(W₁), Z₀} and the corresponding optimal solution W₁ or W′. At this point, the algorithm ends. Otherwise, go to step 5;

Step 5: If n_c > n_T, then output W₁ and Z(W₁). At this point, the algorithm terminates; otherwise, let W₀ = W₁ return to step 2.

Fig. 4

The correlation model and solution scheme between the cause and the fault alarm can be obtained by solving the above method.

Services are ultimately carried on equipment and links. The equipment should include three states: operation, maintenance, and failure. The corresponding power business includes three states: normal, interrupted, and detour. The relationship between the states is shown in Figure 5 below. The overhaul is a known deterministic action, and the business path is migrated before overhaul without affecting the transmission of the business. Because the service interruption and detour caused by equipment or link failure are unknown, failure to handle the interruption in time is likely to cause various security incidents [11]. However, there is currently a lack of adequate analysis for the risks of circuitous state networks. So we will quantify and model the risk level.

Fig. 5

In the power communication network, the business operation risk refers to the possibility of unsafe operation of the power grid due to the failure of the power communication service channel. Therefore, balancing the number of channels for essential services to reduce business operation risks will become a necessary technical means to improve the reliability of power communication network services.

Risk is defined as the product of the probability and the value of the impact after it occurs. Then the risk quantification formula s_i for business risk value E_i is as follows: (11) Ei=Pi⋅Di {E_i} = {P_i} \cdot {D_i} P_i represents the probability of a failure alarm in the event. D_i is the degree of influence corresponding to business E_i. The quantification of D_i is as follows: (12) Di=wi×vih×Ci {D_i} = {w_i} \times {v_{ih}} \times {C_i} v_ih is the weight of category h of business i. w_i is the weighted sum of the equipment and link carried by service s_i. C_i is the business weight. We assume that there are altekrnative routes for service s_i, and P_k each path is denoted as P_k = {V_k, E_k}. Where V_k is the device node on the path. E_k is the link on path k. Then w_i can be further calculated as follows: (13) wi=∑k=1kφk(∑vj∈Vkaj+∑elm∈Ekβlm) {w_i} = \sum\nolimits_{k = 1}^k {\varphi _k}\left( {\sum\nolimits_{{v_j} \in {V_k}} {a_j} + \sum\nolimits_{{e_{lm}} \in {E_k}} {\beta _{lm}}} \right) Among them φ_k is the importance weight of the k-th route. a_j is the weight of the device node V_j in V_k. β_lm is the weight of link elm in E_k (l, m is the device node number).